Project - Plants Seedling Classification

---

Background & Context:

• In recent times, the field of agriculture has been in urgent need of modernizing, since the amount of manual work people need to put in to check if plants are growing correctly is still highly extensive. Despite several advances in agricultural technology, people working in the agricultural industry still need to have the ability to sort and recognize different plants and weeds, which takes a lot of time and effort in the long term.

• The potential is ripe for this trillion-dollar industry to be greatly impacted by technological innovations that cut down on the requirement for manual labor, and this is where Artificial Intelligence can benefit the workers in this field, as the time and energy required to identify plant seedlings will be greatly shortened by the use of AI and Deep Learning. The ability to do so far more efficiently and even more effectively than experienced manual labor could lead to better crop yields, the freeing up of human involvement for higher-order agricultural decision making, and in the long term will result in more sustainable environmental practices in agriculture as well.

Objective:

• The Aarhus University Signal Processing group, in collaboration with the University of Southern Denmark, has provided the data containing images of unique plants belonging to 12 different species.
• The goal of this project is to create a classifier capable of determining a plant's species from an image. by building a Convolutional Neural Network model which would classify the plant seedlings into their respective 12 categories.

Data Description:

This dataset contains images of unique plants belonging to 12 different species. The data file names are:

• images.npy
• Label.csv

Due to the large volume of data, the images were converted to numpy arrays and stored in images.npy file and the corresponding labels are also put into Labels.csv so that you can work on the data/project seamlessly without having to worry about the high data volume.

List of Plant Species

The dataset comprises of 12 plant species.

• Black-grass
• Charlock
• Cleavers
• Common Chickweed
• Common Wheat
• Fat Hen
• Loose Silky-bent
• Maize
• Scentless Mayweed
• Shepherds Purse
• Small-flowered Cranesbill
• Sugar beet

---

from google.colab import drive
drive.mount('/content/drive')

Mounted at /content/drive

Importing required Libraries/Packages

---

import os
import numpy as np                                                                               # Importing numpy for Matrix Operations
import pandas as pd                                                                              # Importing pandas to read CSV files
import matplotlib.pyplot as plt                                                                  # Importting matplotlib for Plotting and visualizing images
import math                                                                                      # Importing math module to perform mathematical operations
import cv2                                                                                       # Importing openCV for image processing
import seaborn as sns                                                                            # Importing seaborn to plot graphs
from sklearn.utils import class_weight

from sklearn import metrics
from sklearn.model_selection import train_test_split, StratifiedKFold, cross_val_score
from sklearn.metrics import classification_report
from tensorflow.keras import backend

from sklearn.metrics import (
    f1_score,
    accuracy_score,
    recall_score,
    precision_score,
    confusion_matrix,
    roc_auc_score,
    plot_confusion_matrix,
    precision_recall_curve,
    roc_curve,
    make_scorer,
)

# Tensorflow modules
from sklearn.preprocessing import LabelBinarizer
import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator                              
from tensorflow.keras.models import Sequential                                                   
from tensorflow.keras.layers import Dense,Dropout,Flatten,Conv2D,MaxPooling2D,BatchNormalization 
from tensorflow.keras.optimizers import Adam,SGD                                                 
from keras import callbacks
from keras.callbacks import ModelCheckpoint
from sklearn import preprocessing                                                                
from sklearn.model_selection import train_test_split                                             
from sklearn.metrics import confusion_matrix                                                     

# Display images using OpenCV
from google.colab.patches import cv2_imshow                                                      

# Ignore warnings
import warnings
warnings.filterwarnings('ignore')

Loading the Dataset

---

# Mount Google drive to access the dataset
from google.colab import drive
drive.mount('/content/drive')

Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount("/content/drive", force_remount=True).

# Loading the image file of the dataset
images = np.load('/content/drive/MyDrive/Colab Notebooks/Project_images.npy')

# Loading the labels file of the dataset
labels = pd.read_csv('/content/drive/MyDrive/Colab Notebooks/Project_Labels.csv')

---

Overview of the dataset

---

• Printing the shape of the images and labels

print(images.shape)
print(labels.shape)

(4750, 128, 128, 3)
(4750, 1)

• There are 4750 RGB images of shape 128 x 128 X 3, each image having 3 channels
• There are also 4750 labels of the plant species corresponding to each images

np.unique(labels)

array(['Black-grass', 'Charlock', 'Cleavers', 'Common Chickweed',
       'Common wheat', 'Fat Hen', 'Loose Silky-bent', 'Maize',
       'Scentless Mayweed', 'Shepherds Purse',
       'Small-flowered Cranesbill', 'Sugar beet'], dtype=object)

• There are 12 unique labels or classes in the dataset

Distribution of each Class

• Printing the unique count of the Classes and the distribution of the images

count=labels.value_counts()
print(count)
print()
print('*'*10)
print()
count=labels.value_counts(normalize=True)
print(count)

Label                    
Loose Silky-bent             654
Common Chickweed             611
Scentless Mayweed            516
Small-flowered Cranesbill    496
Fat Hen                      475
Charlock                     390
Sugar beet                   385
Cleavers                     287
Black-grass                  263
Shepherds Purse              231
Common wheat                 221
Maize                        221
dtype: int64

**********

Label                    
Loose Silky-bent             0.137684
Common Chickweed             0.128632
Scentless Mayweed            0.108632
Small-flowered Cranesbill    0.104421
Fat Hen                      0.100000
Charlock                     0.082105
Sugar beet                   0.081053
Cleavers                     0.060421
Black-grass                  0.055368
Shepherds Purse              0.048632
Common wheat                 0.046526
Maize                        0.046526
dtype: float64

Observations:

• The Class "Loose Silky-bent" is the plant species with more number of images(almost 14%)
• The "Common Wheat" & "Maize" is the least plant species with less number of images. There are only 4.7% ofimages
• We see that the classes are not uniformly distributed and needs to be handled

Plotting few random Images and its corresponding labels

def plot_images(images,labels1):
  num_classes=12                                                                # Number of Classes
  categories=np.unique(labels1)   
  keys=dict(labels1['Label'])                                                   # Obtaing the unique classes from labels
  rows = 5                                                                      # Defining number of rows=5
  cols = 5                                                                      # Defining number of columns=5
  fig = plt.figure(figsize=(20, 15))                                            # Defining the figure size to 20x15
  for i in range(cols):
      for j in range(rows):
          random_index = np.random.randint(0, len(labels1))                     # Generating random indices from the data and plotting the images
          ax = fig.add_subplot(rows, cols, i * rows + j + 1)                    # Adding subplots with 5 rows and 5 columns
          ax.imshow(images[random_index, :])                                    # Plotting the image 
          ax.set_title(keys[random_index])
  plt.show()

plot_images(images,labels)

Treating images based on BGR vs RGB

---

• Checking the images for BGR vs RGB format

cv2_imshow(images[5])

plt.imshow(images[5])

<matplotlib.image.AxesImage at 0x7f3ff4297190>

Observations:

• We can observe that the images are being shown in different colors when plotted with openCV and matplotlib, as OpenCV reads images in BGR format and this shows that the given numpy arrays were generated from the original images using OpenCV.
• We will convert these BGR images to RGB images so we could interpret them easily.

# Converting the images from BGR to RGB using cvtColor function of OpenCV
for i in range(len(images)):
  images[i] = cv2.cvtColor(images[i], cv2.COLOR_BGR2RGB)

• The images were converted to RGB format

Exploratory Data Analysis

---

count=labels.value_counts()
print(count)
print()
print('*'*10)
print()
count=labels.value_counts(normalize=True)
print(count)

Label                    
Loose Silky-bent             654
Common Chickweed             611
Scentless Mayweed            516
Small-flowered Cranesbill    496
Fat Hen                      475
Charlock                     390
Sugar beet                   385
Cleavers                     287
Black-grass                  263
Shepherds Purse              231
Common wheat                 221
Maize                        221
dtype: int64

**********

Label                    
Loose Silky-bent             0.137684
Common Chickweed             0.128632
Scentless Mayweed            0.108632
Small-flowered Cranesbill    0.104421
Fat Hen                      0.100000
Charlock                     0.082105
Sugar beet                   0.081053
Cleavers                     0.060421
Black-grass                  0.055368
Shepherds Purse              0.048632
Common wheat                 0.046526
Maize                        0.046526
dtype: float64

Observations:

• The Class "Loose Silky-bent" is the plant species with more number of images(almost 14%)
• The "Common Wheat" & "Maize" is the least plant species with less number of images. There are only 4.7% ofimages
• We see that the classes are not uniformly distributed and needs to be handled

# Plotting the distribution of the classes
sns.countplot(labels['Label'])
plt.xticks(rotation='vertical')

(array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11]),
 <a list of 12 Text major ticklabel objects>)

Inferences:

• We see that the classes are not uniformly distributed
• The Class "Loose Silky-bent" seems to be having more number of images(around 654) when compared with the "Common Wheat" & "Maize" which has only 221 images.
• We might need to define class weights to balance the classes during training

Summary of Dataset & EDA

• There are 4750 images in RGB format with size dimentsions of 128 x 128
• There are 4720 class labels corresponding to each images inthe image dataset
• The images are in BGR format and needs to be converted to RGB format
• There are 12 unique classes or labels in this dataset
• The Class "Loose Silky-bent" is the plant species with more number of images (almost 14%)
• The "Common Wheat" & "Maize" is the least plant species with less number of images. There are only 4.7% of images
• The classes are not uniformly distributed and needs to be handled

Data Preprocessing

---

Visualizing images using Gaussian Blur

plt.imshow(images[3])

<matplotlib.image.AxesImage at 0x7f3ff1b06f10>

# Applying Gaussian Blur to denoise the images
images_gb=[]
for i in range(len(images)):
  # gb[i] = cv2.cvtColor(images[i], cv2.COLOR_BGR2RGB)
  images_gb.append(cv2.GaussianBlur(images[i], ksize =(3,3),sigmaX =  0))

plt.imshow(images_gb[3])

<matplotlib.image.AxesImage at 0x7f3ff11f2110>

Observations:

• By applying the Gaussion Blur, the edges of the images are more clearly visible while the other portions are blurred. This would help in predicting the image better than the original image

Resizing images

• As the size of the images is large, it may be computationally expensive to train on these larger images; We will reduce the image size from 128 to 64.

images_decreased=[]
height = 64
width = 64
dimensions = (width, height)  
for i in range(len(images)):
  images_decreased.append( cv2.resize(images_gb[i], dimensions, interpolation=cv2.INTER_LINEAR))

# Displaying few images post applying the Gaussian Blur & resizing the shape
plot_images(np.array(images_decreased), labels)

Data Processing for modeling

---

Splitting the dataset

• As we have less images in our dataset, we will only use 10% of our data for testing, 10% of our data for validation and 80% of our data for training.
• We are using the train_test_split() function from scikit-learn. Here, we split the dataset into three parts, train,test and validation.

from sklearn.model_selection import train_test_split
X_temp, X_test, y_temp, y_test = train_test_split(np.array(images_decreased),labels , test_size=0.1, random_state=42,stratify=labels)
X_train, X_val, y_train, y_val = train_test_split(X_temp,y_temp , test_size=0.1, random_state=42,stratify=y_temp)

print(X_train.shape,y_train.shape)
print(X_val.shape,y_val.shape)
print(X_test.shape,y_test.shape)

(3847, 64, 64, 3) (3847, 1)
(428, 64, 64, 3) (428, 1)
(475, 64, 64, 3) (475, 1)

Making the data compatible:

• Encoding Target labels from names to one hot vectors
• Normalizing the values
• Applying the Class weights to balance the Classes

Encoding the target labels

• Converting labels from names to one hot vectors.

# Using Labelbinarizer to do the one hot encoding 

enc = LabelBinarizer()
y_train_encoded = enc.fit_transform(y_train)
y_val_encoded=enc.transform(y_val)
y_test_encoded=enc.transform(y_test)

Data Normalization

• We will normalize the image pixels to range between 0-1 instead of the image pixel ranging from 0-255, by dividing the pixel values by 255

# Normalizing the image pixels
X_train_normalized = X_train.astype('float32')/255.0
X_val_normalized = X_val.astype('float32')/255.0
X_test_normalized = X_test.astype('float32')/255.0

Checking the shape of the data

print(X_train_normalized.shape,y_train_encoded.shape)
print(X_val_normalized.shape,y_val_encoded.shape)
print(X_test_normalized.shape,y_test_encoded.shape)

(3847, 64, 64, 3) (3847, 12)
(428, 64, 64, 3) (428, 12)
(475, 64, 64, 3) (475, 12)

• The shape of the image dataset has been converted to 64 x 64 size

Initializing Class weights to balance the Classes

• The dataset has a imbalance in the class/image distribution and needs to be handled by applying class weights for those that has less number of images when compared with classes that have more number of images

labelList = np.unique(labels)
class_weights = class_weight.compute_class_weight(class_weight = "balanced", classes = np.array(labelList), y = y_train.values.reshape(-1))
class_weights = dict(zip(np.array(range(len(labelList))), class_weights))

#print calculated class weights
class_weights

{0: 1.5050860719874803,
 1: 1.0145042194092826,
 2: 1.3818247126436782,
 3: 0.6476430976430977,
 4: 1.7909683426443204,
 5: 0.8348524305555556,
 6: 0.604874213836478,
 7: 1.7909683426443204,
 8: 0.7669457735247209,
 9: 1.714349376114082,
 10: 0.7974709784411277,
 11: 1.0275106837606838}

• The class_weights variable will be applied to the model as a parameter

Data Preprocessing Summary:

• Gaussian Blur has been applied on the images to reduce the noise and focus on the primary edges of the images
• The images were resized from 128 to 64 size
• The dataset was split into training, validation & test sets with a distribution of 80%, 10% & 10%
• The Target labels were One Hot encoded using Labelbinarizer
• Normalization was applied on the image to convert it to a 0-1 range
• Class weights have been computed to balance the classes that has less distribution of images

We will now apply modeling techniques to determine the best model to identify the plant species

Model Building

---

# defining a function to compute different metrics to check performance of a classification model built using sklearn

def model_performance_classification_sklearn_with_threshold(model, predictors, target, threshold=0.5):

    y_test_pred = model.predict(predictors)
    y_test_pred_classes = np.argmax(y_test_pred,axis=1)
    
    normal_y_test = np.argmax(target,axis=1)

    # Plotting the Confusion Matrix using confusion matrix() function which is also predefined tensorflow module
    confusion_matrix = tf.math.confusion_matrix(normal_y_test,y_test_pred_classes)
    f, ax = plt.subplots(figsize=(10, 8))
    sns.heatmap(
        confusion_matrix,
        annot=True,
        linewidths=.4,
        fmt="d",
        square=True,
        ax=ax
    )
    plt.show()

    acc = accuracy_score(normal_y_test, y_test_pred_classes)
    recall = recall_score(normal_y_test, y_test_pred_classes, average='weighted')  # to compute Recall
    precision = precision_score(normal_y_test, y_test_pred_classes, average='weighted')  # to compute Precision
    f1 = f1_score(normal_y_test, y_test_pred_classes, average='weighted')  # to compute F1-score

    # creating a dataframe of metrics
    df_perf = pd.DataFrame(
        {
            "Accuracy": acc,
            "Recall": recall,
            "Precision": precision,
            "F1": f1,
        },
        index=[0],
    )

    return df_perf

Model 0 - Simple Convolutional Neural Network (CNN)

• Creating a base CNN model

# Clearing backend
backend.clear_session()

# Fixing the seed for random number generators
import random
np.random.seed(42)
random.seed(42)
tf.random.set_seed(42)

# Intializing a sequential model
model = Sequential()

# Adding first conv layer with 64 filters and kernel size 3x3 , padding 'same' provides the output size same as the input size
# Input_shape denotes input image dimension of MNIST images
model.add(Conv2D(126, (3, 3), activation='relu', padding="same", input_shape=(64, 64, 3)))

# Adding max pooling to reduce the size of output of first conv layer
model.add(Conv2D(64, (3, 3), activation='relu', padding="same"))
model.add(MaxPooling2D((2, 2), padding = 'same'))

model.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model.add(MaxPooling2D((2, 2), padding = 'same'))

# flattening the output of the conv layer after max pooling to make it ready for creating dense connections
model.add(Flatten())

# Adding the output layer with 10 neurons and activation functions as softmax since this is a multi-class classification problem  
model.add(Dense(12, activation='softmax'))

# Using Adam Optimizer
opt = Adam()

# Compile model
model.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])

Observations:

• We are using the loss function - categorical_crossentropy as this is a multi-class classification problem and Adam as the optimizer
• Also, we are choosing accuracy as the metric to measure the performance of the model

Fitting the model on the train data

• Applying Epochs & Batch Size parameters and class weights to balance the classes.

history_1 = model.fit(
            X_train_normalized, y_train_encoded,
            epochs=50,
            validation_data=(X_val_normalized,y_val_encoded),
            batch_size=50,
            verbose=2,
            class_weight = class_weights,
)

Epoch 1/50
77/77 - 12s - loss: 2.3389 - accuracy: 0.1469 - val_loss: 1.9997 - val_accuracy: 0.2547 - 12s/epoch - 156ms/step
Epoch 2/50
77/77 - 2s - loss: 1.6928 - accuracy: 0.3824 - val_loss: 1.5727 - val_accuracy: 0.3949 - 2s/epoch - 31ms/step
Epoch 3/50
77/77 - 2s - loss: 1.3897 - accuracy: 0.4903 - val_loss: 1.3451 - val_accuracy: 0.5280 - 2s/epoch - 31ms/step
Epoch 4/50
77/77 - 2s - loss: 1.1807 - accuracy: 0.5745 - val_loss: 1.2738 - val_accuracy: 0.5584 - 2s/epoch - 31ms/step
Epoch 5/50
77/77 - 2s - loss: 1.0721 - accuracy: 0.6181 - val_loss: 1.1801 - val_accuracy: 0.6121 - 2s/epoch - 31ms/step
Epoch 6/50
77/77 - 2s - loss: 0.9368 - accuracy: 0.6769 - val_loss: 1.1836 - val_accuracy: 0.5911 - 2s/epoch - 31ms/step
Epoch 7/50
77/77 - 2s - loss: 0.8073 - accuracy: 0.7208 - val_loss: 1.2029 - val_accuracy: 0.6098 - 2s/epoch - 31ms/step
Epoch 8/50
77/77 - 2s - loss: 0.6621 - accuracy: 0.7746 - val_loss: 1.0675 - val_accuracy: 0.6822 - 2s/epoch - 31ms/step
Epoch 9/50
77/77 - 2s - loss: 0.5488 - accuracy: 0.8089 - val_loss: 1.0082 - val_accuracy: 0.7173 - 2s/epoch - 31ms/step
Epoch 10/50
77/77 - 2s - loss: 0.4649 - accuracy: 0.8459 - val_loss: 1.0116 - val_accuracy: 0.7079 - 2s/epoch - 31ms/step
Epoch 11/50
77/77 - 2s - loss: 0.4349 - accuracy: 0.8518 - val_loss: 1.0349 - val_accuracy: 0.7243 - 2s/epoch - 31ms/step
Epoch 12/50
77/77 - 2s - loss: 0.3803 - accuracy: 0.8695 - val_loss: 1.1288 - val_accuracy: 0.7196 - 2s/epoch - 31ms/step
Epoch 13/50
77/77 - 2s - loss: 0.3313 - accuracy: 0.8869 - val_loss: 1.1797 - val_accuracy: 0.6893 - 2s/epoch - 31ms/step
Epoch 14/50
77/77 - 2s - loss: 0.2580 - accuracy: 0.9124 - val_loss: 1.1924 - val_accuracy: 0.7243 - 2s/epoch - 31ms/step
Epoch 15/50
77/77 - 2s - loss: 0.2173 - accuracy: 0.9194 - val_loss: 1.1676 - val_accuracy: 0.7407 - 2s/epoch - 31ms/step
Epoch 16/50
77/77 - 2s - loss: 0.1867 - accuracy: 0.9374 - val_loss: 1.0888 - val_accuracy: 0.7617 - 2s/epoch - 31ms/step
Epoch 17/50
77/77 - 2s - loss: 0.1501 - accuracy: 0.9504 - val_loss: 1.2759 - val_accuracy: 0.7079 - 2s/epoch - 31ms/step
Epoch 18/50
77/77 - 2s - loss: 0.1152 - accuracy: 0.9620 - val_loss: 1.2993 - val_accuracy: 0.7500 - 2s/epoch - 32ms/step
Epoch 19/50
77/77 - 2s - loss: 0.1190 - accuracy: 0.9548 - val_loss: 1.4037 - val_accuracy: 0.7360 - 2s/epoch - 31ms/step
Epoch 20/50
77/77 - 2s - loss: 0.0807 - accuracy: 0.9745 - val_loss: 1.3861 - val_accuracy: 0.7290 - 2s/epoch - 31ms/step
Epoch 21/50
77/77 - 2s - loss: 0.0556 - accuracy: 0.9826 - val_loss: 1.4163 - val_accuracy: 0.7383 - 2s/epoch - 32ms/step
Epoch 22/50
77/77 - 2s - loss: 0.0504 - accuracy: 0.9839 - val_loss: 1.5387 - val_accuracy: 0.7103 - 2s/epoch - 31ms/step
Epoch 23/50
77/77 - 2s - loss: 0.0575 - accuracy: 0.9800 - val_loss: 1.7358 - val_accuracy: 0.6986 - 2s/epoch - 32ms/step
Epoch 24/50
77/77 - 2s - loss: 0.0645 - accuracy: 0.9766 - val_loss: 1.5966 - val_accuracy: 0.7430 - 2s/epoch - 31ms/step
Epoch 25/50
77/77 - 2s - loss: 0.0560 - accuracy: 0.9870 - val_loss: 1.8412 - val_accuracy: 0.6846 - 2s/epoch - 31ms/step
Epoch 26/50
77/77 - 2s - loss: 0.0689 - accuracy: 0.9769 - val_loss: 1.6242 - val_accuracy: 0.7500 - 2s/epoch - 31ms/step
Epoch 27/50
77/77 - 2s - loss: 0.0299 - accuracy: 0.9919 - val_loss: 1.7098 - val_accuracy: 0.7383 - 2s/epoch - 31ms/step
Epoch 28/50
77/77 - 2s - loss: 0.0628 - accuracy: 0.9792 - val_loss: 2.0160 - val_accuracy: 0.6869 - 2s/epoch - 31ms/step
Epoch 29/50
77/77 - 2s - loss: 0.1412 - accuracy: 0.9504 - val_loss: 1.7026 - val_accuracy: 0.6963 - 2s/epoch - 31ms/step
Epoch 30/50
77/77 - 2s - loss: 0.0348 - accuracy: 0.9896 - val_loss: 1.6733 - val_accuracy: 0.7383 - 2s/epoch - 32ms/step
Epoch 31/50
77/77 - 2s - loss: 0.0121 - accuracy: 0.9982 - val_loss: 1.7460 - val_accuracy: 0.7500 - 2s/epoch - 31ms/step
Epoch 32/50
77/77 - 2s - loss: 0.0059 - accuracy: 0.9995 - val_loss: 1.8167 - val_accuracy: 0.7477 - 2s/epoch - 32ms/step
Epoch 33/50
77/77 - 2s - loss: 0.0038 - accuracy: 1.0000 - val_loss: 1.8397 - val_accuracy: 0.7453 - 2s/epoch - 31ms/step
Epoch 34/50
77/77 - 2s - loss: 0.0031 - accuracy: 1.0000 - val_loss: 1.8952 - val_accuracy: 0.7500 - 2s/epoch - 31ms/step
Epoch 35/50
77/77 - 2s - loss: 0.0024 - accuracy: 1.0000 - val_loss: 1.9336 - val_accuracy: 0.7453 - 2s/epoch - 32ms/step
Epoch 36/50
77/77 - 2s - loss: 0.0022 - accuracy: 1.0000 - val_loss: 1.9575 - val_accuracy: 0.7453 - 2s/epoch - 32ms/step
Epoch 37/50
77/77 - 3s - loss: 0.0018 - accuracy: 1.0000 - val_loss: 1.9808 - val_accuracy: 0.7453 - 3s/epoch - 33ms/step
Epoch 38/50
77/77 - 2s - loss: 0.0016 - accuracy: 1.0000 - val_loss: 2.0010 - val_accuracy: 0.7453 - 2s/epoch - 32ms/step
Epoch 39/50
77/77 - 2s - loss: 0.0015 - accuracy: 1.0000 - val_loss: 2.0380 - val_accuracy: 0.7453 - 2s/epoch - 31ms/step
Epoch 40/50
77/77 - 2s - loss: 0.0013 - accuracy: 1.0000 - val_loss: 2.0738 - val_accuracy: 0.7453 - 2s/epoch - 32ms/step
Epoch 41/50
77/77 - 2s - loss: 0.0011 - accuracy: 1.0000 - val_loss: 2.0776 - val_accuracy: 0.7453 - 2s/epoch - 32ms/step
Epoch 42/50
77/77 - 2s - loss: 0.0010 - accuracy: 1.0000 - val_loss: 2.1199 - val_accuracy: 0.7477 - 2s/epoch - 32ms/step
Epoch 43/50
77/77 - 2s - loss: 9.3207e-04 - accuracy: 1.0000 - val_loss: 2.1357 - val_accuracy: 0.7453 - 2s/epoch - 32ms/step
Epoch 44/50
77/77 - 2s - loss: 8.6076e-04 - accuracy: 1.0000 - val_loss: 2.1517 - val_accuracy: 0.7477 - 2s/epoch - 32ms/step
Epoch 45/50
77/77 - 2s - loss: 7.8639e-04 - accuracy: 1.0000 - val_loss: 2.1800 - val_accuracy: 0.7477 - 2s/epoch - 32ms/step
Epoch 46/50
77/77 - 2s - loss: 7.3427e-04 - accuracy: 1.0000 - val_loss: 2.1900 - val_accuracy: 0.7477 - 2s/epoch - 32ms/step
Epoch 47/50
77/77 - 2s - loss: 6.7935e-04 - accuracy: 1.0000 - val_loss: 2.2172 - val_accuracy: 0.7477 - 2s/epoch - 32ms/step
Epoch 48/50
77/77 - 2s - loss: 6.4884e-04 - accuracy: 1.0000 - val_loss: 2.2558 - val_accuracy: 0.7500 - 2s/epoch - 31ms/step
Epoch 49/50
77/77 - 2s - loss: 5.8926e-04 - accuracy: 1.0000 - val_loss: 2.2428 - val_accuracy: 0.7477 - 2s/epoch - 32ms/step
Epoch 50/50
77/77 - 2s - loss: 5.4974e-04 - accuracy: 1.0000 - val_loss: 2.2728 - val_accuracy: 0.7453 - 2s/epoch - 32ms/step

Model Evaluation

plt.plot(history_1.history['accuracy'])
plt.plot(history_1.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()

Observations:

• We can see from the above plot that the training accuracy is overfitting. Also, the validation accuracy is over fitting with the Training accuracy.
• We might need to apply other models to tune it further by applying BatchNormalazation, Pooling and dropout parameters

Model Evaluation on Training dataset

model.evaluate(X_train_normalized, y_train_encoded, verbose=1)
model0_TrainDF = model_performance_classification_sklearn_with_threshold(
    model, X_train_normalized, y_train_encoded,
)
model0_TrainDF

121/121 [==============================] - 1s 10ms/step - loss: 5.0907e-04 - accuracy: 1.0000
121/121 [==============================] - 1s 7ms/step

	Accuracy	Recall	Precision	F1
0	1.0	1.0	1.0	1.0

Observations

• The Training accuracy of this model is 100% and the other metrics are also at 100% resulting in overfitting
• All the classes are well classified and probably overfitting

Model Evaluation on Validation dataset

model.evaluate(X_val_normalized, y_val_encoded, verbose=1)
model0_ValDF = model_performance_classification_sklearn_with_threshold(
    model, X_val_normalized,y_val_encoded,
)
model0_ValDF

14/14 [==============================] - 0s 19ms/step - loss: 2.2728 - accuracy: 0.7453
14/14 [==============================] - 0s 7ms/step

	Accuracy	Recall	Precision	F1
0	0.745327	0.745327	0.736978	0.738901

Observations

• The Validation accuracy of this model is 74.5% and the other metrics also are in the same range. Precision has a higher metric
• When compared with the Training accurace, there is ~26% variance in the Accuracy scores. There is room for improvement to reduce this variance

Model Evaluation on Test dataset

model.evaluate(X_test_normalized, y_test_encoded, verbose=2)
model0_TestDF = model_performance_classification_sklearn_with_threshold(model, X_test_normalized, y_test_encoded)
model0_TestDF

15/15 - 0s - loss: 2.4591 - accuracy: 0.7326 - 317ms/epoch - 21ms/step
15/15 [==============================] - 0s 7ms/step

	Accuracy	Recall	Precision	F1
0	0.732632	0.732632	0.729495	0.72711

Observations:

• The Testing accuracy of this model is 73.3% and the other metrics also are in the same range. Precision has a higher metric
• When compared with the Training accuracy (100%), the Testing accuracy has a big gap of almsot 27% variance
• Using the classification chart, we can observe that the classes are not predicted correctly. The classes are incorrectly classified
• We need to tune the model with additional layers along with Pooling, Batch Normalization and dropouts as applicable. We will apply these parameters in the next model.

Visualizing the Predictions:

# Visualizing the predicted and correct label of images from test data
rows = 1                                                                        
cols = 5                                                                        
for i in range(cols):
    for j in range(rows):
      random_index = np.random.randint(0, 400)    
      plt.imshow(X_test[random_index])
      plt.show()

      print('Predicted Label', enc.inverse_transform(model.predict((X_test_normalized[random_index].reshape(1,64,64,3)))))   
      print('True Label', enc.inverse_transform(y_test_encoded)[random_index])                                               

1/1 [==============================] - 0s 103ms/step
Predicted Label ['Sugar beet']
True Label Sugar beet

1/1 [==============================] - 0s 16ms/step
Predicted Label ['Loose Silky-bent']
True Label Loose Silky-bent

1/1 [==============================] - 0s 16ms/step
Predicted Label ['Loose Silky-bent']
True Label Loose Silky-bent

1/1 [==============================] - 0s 17ms/step
Predicted Label ['Scentless Mayweed']
True Label Scentless Mayweed

1/1 [==============================] - 0s 22ms/step
Predicted Label ['Small-flowered Cranesbill']
True Label Small-flowered Cranesbill

Inference:

• The random images selected has the correct prediction so far in this model.

---

Model 1 - Convolutional Neural Network (CNN) with addiitonal Layers

• Creating a CNN model with addiitonal layers & tuning by using MaxPooling, Batch Normalization & Droputs.

# Clearing backend
backend.clear_session()

# Fixing the seed for random number generators
import random
np.random.seed(42)
random.seed(42)
tf.random.set_seed(42)

# Intializing a sequential model
model = Sequential()

# Adding first conv layer with 64 filters and kernel size 3x3 , padding 'same' provides the output size same as the input size
# Input_shape denotes input image dimension of images
model.add(Conv2D(64, (3, 3), activation='relu', padding="same", input_shape=(64, 64, 3)))
# Adding max pooling to reduce the size of output of first conv layer
model.add(MaxPooling2D(2, 2))
model.add(BatchNormalization())

model.add(Conv2D(64, (3, 3), activation='relu', padding="same"))
model.add(BatchNormalization())
model.add(MaxPooling2D(2, 2))
model.add(Dense(32, activation='relu'))

model.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model.add(MaxPooling2D(2, 2))
model.add(Dropout(0.3))

model.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model.add(Dense(32, activation='relu'))
model.add(MaxPooling2D(2, 2))
model.add(Dropout(0.3))

model.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model.add(MaxPooling2D(2, 2))

model.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model.add(MaxPooling2D(2, 2))
model.add(BatchNormalization())

# flattening the output of the conv layer after max pooling to make it ready for creating dense connections
model.add(Flatten())

# Adding a fully connected dense layer with 100 neurons    
model.add(Dense(64, activation='relu'))
model.add(Dropout(0.3))

# Adding a fully connected dense layer with 100 neurons    
model.add(Dense(32, activation='relu'))
model.add(Dropout(0.3))

model.add(Dense(32, activation='relu'))

# Adding the output layer with 12 neurons and activation functions as softmax since this is a multi-class classification problem  
model.add(Dense(12, activation='softmax'))

# Using Adam Optimizer
opt=Adam()

# Compile model
model.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])

# Generating the summary of the model
model.summary()

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d (Conv2D)             (None, 64, 64, 64)        1792      
                                                                 
 max_pooling2d (MaxPooling2D  (None, 32, 32, 64)       0         
 )                                                               
                                                                 
 batch_normalization (BatchN  (None, 32, 32, 64)       256       
 ormalization)                                                   
                                                                 
 conv2d_1 (Conv2D)           (None, 32, 32, 64)        36928     
                                                                 
 batch_normalization_1 (Batc  (None, 32, 32, 64)       256       
 hNormalization)                                                 
                                                                 
 max_pooling2d_1 (MaxPooling  (None, 16, 16, 64)       0         
 2D)                                                             
                                                                 
 dense (Dense)               (None, 16, 16, 32)        2080      
                                                                 
 conv2d_2 (Conv2D)           (None, 16, 16, 32)        9248      
                                                                 
 max_pooling2d_2 (MaxPooling  (None, 8, 8, 32)         0         
 2D)                                                             
                                                                 
 dropout (Dropout)           (None, 8, 8, 32)          0         
                                                                 
 conv2d_3 (Conv2D)           (None, 8, 8, 32)          9248      
                                                                 
 dense_1 (Dense)             (None, 8, 8, 32)          1056      
                                                                 
 max_pooling2d_3 (MaxPooling  (None, 4, 4, 32)         0         
 2D)                                                             
                                                                 
 dropout_1 (Dropout)         (None, 4, 4, 32)          0         
                                                                 
 conv2d_4 (Conv2D)           (None, 4, 4, 32)          9248      
                                                                 
 max_pooling2d_4 (MaxPooling  (None, 2, 2, 32)         0         
 2D)                                                             
                                                                 
 conv2d_5 (Conv2D)           (None, 2, 2, 32)          9248      
                                                                 
 max_pooling2d_5 (MaxPooling  (None, 1, 1, 32)         0         
 2D)                                                             
                                                                 
 batch_normalization_2 (Batc  (None, 1, 1, 32)         128       
 hNormalization)                                                 
                                                                 
 flatten (Flatten)           (None, 32)                0         
                                                                 
 dense_2 (Dense)             (None, 64)                2112      
                                                                 
 dropout_2 (Dropout)         (None, 64)                0         
                                                                 
 dense_3 (Dense)             (None, 32)                2080      
                                                                 
 dropout_3 (Dropout)         (None, 32)                0         
                                                                 
 dense_4 (Dense)             (None, 32)                1056      
                                                                 
 dense_5 (Dense)             (None, 12)                396       
                                                                 
=================================================================
Total params: 85,132
Trainable params: 84,812
Non-trainable params: 320
_________________________________________________________________

Observations:

• There are a total of 85,132 params with 84,812 trainable params
• We have applied Convolution along with BatchNormalization, pooling and droputs to tune the model along with the flat layer
• We are using the loss function - categorical_crossentropy as this is a multi-class classification problem and Adam as the optimizer
• Also, we are choosing accuracy as the metric to measure the performance of the model

Fitting the model on the train data

• Applying Epochs & Batch Size parameters and class weights to balance the classes.

history_1 = model.fit(
            X_train_normalized, y_train_encoded,
            epochs=200,
            validation_data=(X_val_normalized,y_val_encoded),
            batch_size=50,
            verbose=2,
            class_weight = class_weights,
)

Epoch 1/200
77/77 - 4s - loss: 2.5024 - accuracy: 0.0876 - val_loss: 2.4939 - val_accuracy: 0.0467 - 4s/epoch - 48ms/step
Epoch 2/200
77/77 - 1s - loss: 2.3346 - accuracy: 0.1432 - val_loss: 2.8157 - val_accuracy: 0.1332 - 1s/epoch - 15ms/step
Epoch 3/200
77/77 - 1s - loss: 1.9805 - accuracy: 0.2441 - val_loss: 4.1275 - val_accuracy: 0.1379 - 1s/epoch - 15ms/step
Epoch 4/200
77/77 - 1s - loss: 1.7452 - accuracy: 0.3091 - val_loss: 3.0697 - val_accuracy: 0.0654 - 1s/epoch - 15ms/step
Epoch 5/200
77/77 - 1s - loss: 1.6754 - accuracy: 0.3442 - val_loss: 4.2307 - val_accuracy: 0.0584 - 1s/epoch - 15ms/step
Epoch 6/200
77/77 - 1s - loss: 1.5965 - accuracy: 0.3717 - val_loss: 2.8036 - val_accuracy: 0.1939 - 1s/epoch - 15ms/step
Epoch 7/200
77/77 - 1s - loss: 1.5088 - accuracy: 0.4047 - val_loss: 2.0689 - val_accuracy: 0.2921 - 1s/epoch - 15ms/step
Epoch 8/200
77/77 - 1s - loss: 1.4395 - accuracy: 0.4409 - val_loss: 1.6932 - val_accuracy: 0.3785 - 1s/epoch - 15ms/step
Epoch 9/200
77/77 - 1s - loss: 1.4033 - accuracy: 0.4435 - val_loss: 4.4969 - val_accuracy: 0.0794 - 1s/epoch - 15ms/step
Epoch 10/200
77/77 - 1s - loss: 1.3644 - accuracy: 0.4549 - val_loss: 2.2258 - val_accuracy: 0.3037 - 1s/epoch - 15ms/step
Epoch 11/200
77/77 - 1s - loss: 1.3183 - accuracy: 0.4760 - val_loss: 1.6239 - val_accuracy: 0.4252 - 1s/epoch - 15ms/step
Epoch 12/200
77/77 - 1s - loss: 1.2908 - accuracy: 0.4936 - val_loss: 2.6793 - val_accuracy: 0.2220 - 1s/epoch - 15ms/step
Epoch 13/200
77/77 - 1s - loss: 1.2478 - accuracy: 0.5188 - val_loss: 1.8548 - val_accuracy: 0.4322 - 1s/epoch - 15ms/step
Epoch 14/200
77/77 - 1s - loss: 1.1699 - accuracy: 0.5537 - val_loss: 1.9590 - val_accuracy: 0.3551 - 1s/epoch - 15ms/step
Epoch 15/200
77/77 - 1s - loss: 1.1976 - accuracy: 0.5532 - val_loss: 1.0908 - val_accuracy: 0.5561 - 1s/epoch - 15ms/step
Epoch 16/200
77/77 - 1s - loss: 1.1502 - accuracy: 0.5623 - val_loss: 1.2331 - val_accuracy: 0.5257 - 1s/epoch - 16ms/step
Epoch 17/200
77/77 - 1s - loss: 1.1101 - accuracy: 0.5711 - val_loss: 1.2543 - val_accuracy: 0.4977 - 1s/epoch - 15ms/step
Epoch 18/200
77/77 - 1s - loss: 1.1053 - accuracy: 0.5703 - val_loss: 1.1487 - val_accuracy: 0.6028 - 1s/epoch - 15ms/step
Epoch 19/200
77/77 - 1s - loss: 1.0982 - accuracy: 0.5950 - val_loss: 2.0382 - val_accuracy: 0.4042 - 1s/epoch - 15ms/step
Epoch 20/200
77/77 - 1s - loss: 1.0326 - accuracy: 0.6218 - val_loss: 1.7301 - val_accuracy: 0.4579 - 1s/epoch - 15ms/step
Epoch 21/200
77/77 - 1s - loss: 1.0022 - accuracy: 0.6369 - val_loss: 0.9595 - val_accuracy: 0.6612 - 1s/epoch - 15ms/step
Epoch 22/200
77/77 - 1s - loss: 0.9375 - accuracy: 0.6356 - val_loss: 1.2293 - val_accuracy: 0.5911 - 1s/epoch - 15ms/step
Epoch 23/200
77/77 - 1s - loss: 0.9104 - accuracy: 0.6551 - val_loss: 1.2346 - val_accuracy: 0.5584 - 1s/epoch - 15ms/step
Epoch 24/200
77/77 - 1s - loss: 0.9008 - accuracy: 0.6579 - val_loss: 0.8256 - val_accuracy: 0.7430 - 1s/epoch - 15ms/step
Epoch 25/200
77/77 - 1s - loss: 0.8675 - accuracy: 0.6795 - val_loss: 1.7327 - val_accuracy: 0.5491 - 1s/epoch - 15ms/step
Epoch 26/200
77/77 - 1s - loss: 0.8419 - accuracy: 0.6800 - val_loss: 0.8968 - val_accuracy: 0.7079 - 1s/epoch - 15ms/step
Epoch 27/200
77/77 - 1s - loss: 0.8221 - accuracy: 0.7024 - val_loss: 1.2312 - val_accuracy: 0.6215 - 1s/epoch - 15ms/step
Epoch 28/200
77/77 - 1s - loss: 0.8289 - accuracy: 0.6953 - val_loss: 1.2880 - val_accuracy: 0.6238 - 1s/epoch - 15ms/step
Epoch 29/200
77/77 - 1s - loss: 0.7696 - accuracy: 0.7070 - val_loss: 1.4673 - val_accuracy: 0.5818 - 1s/epoch - 16ms/step
Epoch 30/200
77/77 - 1s - loss: 0.7763 - accuracy: 0.7224 - val_loss: 1.4805 - val_accuracy: 0.5701 - 1s/epoch - 15ms/step
Epoch 31/200
77/77 - 1s - loss: 0.7190 - accuracy: 0.7289 - val_loss: 1.0720 - val_accuracy: 0.5935 - 1s/epoch - 15ms/step
Epoch 32/200
77/77 - 1s - loss: 0.7098 - accuracy: 0.7393 - val_loss: 0.7571 - val_accuracy: 0.7850 - 1s/epoch - 15ms/step
Epoch 33/200
77/77 - 1s - loss: 0.7107 - accuracy: 0.7234 - val_loss: 1.0457 - val_accuracy: 0.6939 - 1s/epoch - 15ms/step
Epoch 34/200
77/77 - 1s - loss: 0.6882 - accuracy: 0.7421 - val_loss: 0.7006 - val_accuracy: 0.7383 - 1s/epoch - 15ms/step
Epoch 35/200
77/77 - 1s - loss: 0.6489 - accuracy: 0.7463 - val_loss: 1.3067 - val_accuracy: 0.6028 - 1s/epoch - 15ms/step
Epoch 36/200
77/77 - 1s - loss: 0.6466 - accuracy: 0.7554 - val_loss: 1.4088 - val_accuracy: 0.6098 - 1s/epoch - 15ms/step
Epoch 37/200
77/77 - 1s - loss: 0.6449 - accuracy: 0.7702 - val_loss: 0.9264 - val_accuracy: 0.6752 - 1s/epoch - 15ms/step
Epoch 38/200
77/77 - 1s - loss: 0.6389 - accuracy: 0.7668 - val_loss: 1.4368 - val_accuracy: 0.5911 - 1s/epoch - 15ms/step
Epoch 39/200
77/77 - 1s - loss: 0.5832 - accuracy: 0.7842 - val_loss: 1.7026 - val_accuracy: 0.5444 - 1s/epoch - 15ms/step
Epoch 40/200
77/77 - 1s - loss: 0.5917 - accuracy: 0.7814 - val_loss: 0.7423 - val_accuracy: 0.7664 - 1s/epoch - 15ms/step
Epoch 41/200
77/77 - 1s - loss: 0.5714 - accuracy: 0.7881 - val_loss: 1.3741 - val_accuracy: 0.5514 - 1s/epoch - 15ms/step
Epoch 42/200
77/77 - 1s - loss: 0.5857 - accuracy: 0.7790 - val_loss: 1.8612 - val_accuracy: 0.5023 - 1s/epoch - 15ms/step
Epoch 43/200
77/77 - 1s - loss: 0.5926 - accuracy: 0.7819 - val_loss: 0.7455 - val_accuracy: 0.7617 - 1s/epoch - 15ms/step
Epoch 44/200
77/77 - 1s - loss: 0.5561 - accuracy: 0.8022 - val_loss: 1.0182 - val_accuracy: 0.7290 - 1s/epoch - 15ms/step
Epoch 45/200
77/77 - 1s - loss: 0.5458 - accuracy: 0.7985 - val_loss: 0.8161 - val_accuracy: 0.6986 - 1s/epoch - 15ms/step
Epoch 46/200
77/77 - 1s - loss: 0.5195 - accuracy: 0.8089 - val_loss: 0.8260 - val_accuracy: 0.7453 - 1s/epoch - 15ms/step
Epoch 47/200
77/77 - 1s - loss: 0.5361 - accuracy: 0.8040 - val_loss: 0.6681 - val_accuracy: 0.8061 - 1s/epoch - 15ms/step
Epoch 48/200
77/77 - 1s - loss: 0.4772 - accuracy: 0.8225 - val_loss: 0.7559 - val_accuracy: 0.7967 - 1s/epoch - 15ms/step
Epoch 49/200
77/77 - 1s - loss: 0.4931 - accuracy: 0.8193 - val_loss: 1.0998 - val_accuracy: 0.6822 - 1s/epoch - 15ms/step
Epoch 50/200
77/77 - 1s - loss: 0.5156 - accuracy: 0.8121 - val_loss: 0.6803 - val_accuracy: 0.7804 - 1s/epoch - 15ms/step
Epoch 51/200
77/77 - 1s - loss: 0.4929 - accuracy: 0.8128 - val_loss: 1.1960 - val_accuracy: 0.6752 - 1s/epoch - 15ms/step
Epoch 52/200
77/77 - 1s - loss: 0.4695 - accuracy: 0.8274 - val_loss: 1.7797 - val_accuracy: 0.6168 - 1s/epoch - 15ms/step
Epoch 53/200
77/77 - 1s - loss: 0.4665 - accuracy: 0.8284 - val_loss: 0.8017 - val_accuracy: 0.7547 - 1s/epoch - 15ms/step
Epoch 54/200
77/77 - 1s - loss: 0.4580 - accuracy: 0.8209 - val_loss: 0.7190 - val_accuracy: 0.7874 - 1s/epoch - 15ms/step
Epoch 55/200
77/77 - 1s - loss: 0.4599 - accuracy: 0.8282 - val_loss: 0.9497 - val_accuracy: 0.7220 - 1s/epoch - 15ms/step
Epoch 56/200
77/77 - 1s - loss: 0.4442 - accuracy: 0.8404 - val_loss: 1.7456 - val_accuracy: 0.5654 - 1s/epoch - 15ms/step
Epoch 57/200
77/77 - 1s - loss: 0.4515 - accuracy: 0.8329 - val_loss: 0.8344 - val_accuracy: 0.7383 - 1s/epoch - 15ms/step
Epoch 58/200
77/77 - 1s - loss: 0.4519 - accuracy: 0.8420 - val_loss: 0.8893 - val_accuracy: 0.7220 - 1s/epoch - 15ms/step
Epoch 59/200
77/77 - 1s - loss: 0.4239 - accuracy: 0.8469 - val_loss: 1.9049 - val_accuracy: 0.4930 - 1s/epoch - 15ms/step
Epoch 60/200
77/77 - 1s - loss: 0.4422 - accuracy: 0.8342 - val_loss: 1.7239 - val_accuracy: 0.6472 - 1s/epoch - 15ms/step
Epoch 61/200
77/77 - 1s - loss: 0.4111 - accuracy: 0.8495 - val_loss: 0.8014 - val_accuracy: 0.7617 - 1s/epoch - 15ms/step
Epoch 62/200
77/77 - 1s - loss: 0.4278 - accuracy: 0.8425 - val_loss: 0.6305 - val_accuracy: 0.7991 - 1s/epoch - 15ms/step
Epoch 63/200
77/77 - 1s - loss: 0.4498 - accuracy: 0.8427 - val_loss: 1.1664 - val_accuracy: 0.6986 - 1s/epoch - 15ms/step
Epoch 64/200
77/77 - 1s - loss: 0.4274 - accuracy: 0.8485 - val_loss: 1.4136 - val_accuracy: 0.5981 - 1s/epoch - 15ms/step
Epoch 65/200
77/77 - 1s - loss: 0.4075 - accuracy: 0.8586 - val_loss: 0.6253 - val_accuracy: 0.8248 - 1s/epoch - 15ms/step
Epoch 66/200
77/77 - 1s - loss: 0.4347 - accuracy: 0.8537 - val_loss: 1.9132 - val_accuracy: 0.5584 - 1s/epoch - 15ms/step
Epoch 67/200
77/77 - 1s - loss: 0.3731 - accuracy: 0.8594 - val_loss: 1.2162 - val_accuracy: 0.6986 - 1s/epoch - 15ms/step
Epoch 68/200
77/77 - 1s - loss: 0.3904 - accuracy: 0.8591 - val_loss: 0.7103 - val_accuracy: 0.7874 - 1s/epoch - 15ms/step
Epoch 69/200
77/77 - 1s - loss: 0.3966 - accuracy: 0.8602 - val_loss: 1.4731 - val_accuracy: 0.6215 - 1s/epoch - 15ms/step
Epoch 70/200
77/77 - 1s - loss: 0.3980 - accuracy: 0.8602 - val_loss: 0.6154 - val_accuracy: 0.8505 - 1s/epoch - 15ms/step
Epoch 71/200
77/77 - 1s - loss: 0.3915 - accuracy: 0.8628 - val_loss: 0.6897 - val_accuracy: 0.8084 - 1s/epoch - 15ms/step
Epoch 72/200
77/77 - 1s - loss: 0.3548 - accuracy: 0.8656 - val_loss: 0.6817 - val_accuracy: 0.8107 - 1s/epoch - 15ms/step
Epoch 73/200
77/77 - 1s - loss: 0.3566 - accuracy: 0.8721 - val_loss: 1.6673 - val_accuracy: 0.5771 - 1s/epoch - 15ms/step
Epoch 74/200
77/77 - 1s - loss: 0.3782 - accuracy: 0.8602 - val_loss: 0.6042 - val_accuracy: 0.8271 - 1s/epoch - 15ms/step
Epoch 75/200
77/77 - 1s - loss: 0.3679 - accuracy: 0.8705 - val_loss: 1.2240 - val_accuracy: 0.6472 - 1s/epoch - 15ms/step
Epoch 76/200
77/77 - 1s - loss: 0.3430 - accuracy: 0.8789 - val_loss: 0.7519 - val_accuracy: 0.7827 - 1s/epoch - 15ms/step
Epoch 77/200
77/77 - 1s - loss: 0.3494 - accuracy: 0.8765 - val_loss: 1.9834 - val_accuracy: 0.5537 - 1s/epoch - 15ms/step
Epoch 78/200
77/77 - 1s - loss: 0.3077 - accuracy: 0.8867 - val_loss: 0.9887 - val_accuracy: 0.7383 - 1s/epoch - 15ms/step
Epoch 79/200
77/77 - 1s - loss: 0.3367 - accuracy: 0.8812 - val_loss: 1.3043 - val_accuracy: 0.6355 - 1s/epoch - 15ms/step
Epoch 80/200
77/77 - 1s - loss: 0.3372 - accuracy: 0.8752 - val_loss: 0.6806 - val_accuracy: 0.8154 - 1s/epoch - 15ms/step
Epoch 81/200
77/77 - 1s - loss: 0.3050 - accuracy: 0.8898 - val_loss: 0.6316 - val_accuracy: 0.8388 - 1s/epoch - 15ms/step
Epoch 82/200
77/77 - 1s - loss: 0.3155 - accuracy: 0.8838 - val_loss: 0.5322 - val_accuracy: 0.8411 - 1s/epoch - 15ms/step
Epoch 83/200
77/77 - 1s - loss: 0.3390 - accuracy: 0.8791 - val_loss: 1.0878 - val_accuracy: 0.7640 - 1s/epoch - 15ms/step
Epoch 84/200
77/77 - 1s - loss: 0.3305 - accuracy: 0.8802 - val_loss: 0.6812 - val_accuracy: 0.8037 - 1s/epoch - 15ms/step
Epoch 85/200
77/77 - 1s - loss: 0.3236 - accuracy: 0.8828 - val_loss: 4.2939 - val_accuracy: 0.4019 - 1s/epoch - 15ms/step
Epoch 86/200
77/77 - 1s - loss: 0.3104 - accuracy: 0.8841 - val_loss: 0.7972 - val_accuracy: 0.8014 - 1s/epoch - 15ms/step
Epoch 87/200
77/77 - 1s - loss: 0.3623 - accuracy: 0.8768 - val_loss: 3.0438 - val_accuracy: 0.4836 - 1s/epoch - 15ms/step
Epoch 88/200
77/77 - 1s - loss: 0.2966 - accuracy: 0.8986 - val_loss: 0.7261 - val_accuracy: 0.7967 - 1s/epoch - 15ms/step
Epoch 89/200
77/77 - 1s - loss: 0.2962 - accuracy: 0.8963 - val_loss: 1.0824 - val_accuracy: 0.7196 - 1s/epoch - 15ms/step
Epoch 90/200
77/77 - 1s - loss: 0.2904 - accuracy: 0.8960 - val_loss: 0.6361 - val_accuracy: 0.8037 - 1s/epoch - 15ms/step
Epoch 91/200
77/77 - 1s - loss: 0.2874 - accuracy: 0.8999 - val_loss: 0.8551 - val_accuracy: 0.7593 - 1s/epoch - 15ms/step
Epoch 92/200
77/77 - 1s - loss: 0.2576 - accuracy: 0.9051 - val_loss: 1.0756 - val_accuracy: 0.7243 - 1s/epoch - 15ms/step
Epoch 93/200
77/77 - 1s - loss: 0.2972 - accuracy: 0.9030 - val_loss: 1.5211 - val_accuracy: 0.6636 - 1s/epoch - 15ms/step
Epoch 94/200
77/77 - 1s - loss: 0.2781 - accuracy: 0.8997 - val_loss: 1.7961 - val_accuracy: 0.6238 - 1s/epoch - 15ms/step
Epoch 95/200
77/77 - 1s - loss: 0.2750 - accuracy: 0.9038 - val_loss: 1.7117 - val_accuracy: 0.6215 - 1s/epoch - 15ms/step
Epoch 96/200
77/77 - 1s - loss: 0.2730 - accuracy: 0.9085 - val_loss: 0.6396 - val_accuracy: 0.8154 - 1s/epoch - 15ms/step
Epoch 97/200
77/77 - 1s - loss: 0.2839 - accuracy: 0.8991 - val_loss: 0.7160 - val_accuracy: 0.8248 - 1s/epoch - 15ms/step
Epoch 98/200
77/77 - 1s - loss: 0.2655 - accuracy: 0.9114 - val_loss: 0.9428 - val_accuracy: 0.7687 - 1s/epoch - 15ms/step
Epoch 99/200
77/77 - 1s - loss: 0.2503 - accuracy: 0.9127 - val_loss: 0.7036 - val_accuracy: 0.8248 - 1s/epoch - 15ms/step
Epoch 100/200
77/77 - 1s - loss: 0.2886 - accuracy: 0.9015 - val_loss: 0.5662 - val_accuracy: 0.8621 - 1s/epoch - 15ms/step
Epoch 101/200
77/77 - 1s - loss: 0.2946 - accuracy: 0.9012 - val_loss: 0.7684 - val_accuracy: 0.7921 - 1s/epoch - 15ms/step
Epoch 102/200
77/77 - 1s - loss: 0.2771 - accuracy: 0.9072 - val_loss: 1.5780 - val_accuracy: 0.6285 - 1s/epoch - 15ms/step
Epoch 103/200
77/77 - 1s - loss: 0.2671 - accuracy: 0.9067 - val_loss: 1.3254 - val_accuracy: 0.7173 - 1s/epoch - 15ms/step
Epoch 104/200
77/77 - 1s - loss: 0.2800 - accuracy: 0.9051 - val_loss: 0.7779 - val_accuracy: 0.7991 - 1s/epoch - 15ms/step
Epoch 105/200
77/77 - 1s - loss: 0.2565 - accuracy: 0.9127 - val_loss: 0.6432 - val_accuracy: 0.8318 - 1s/epoch - 15ms/step
Epoch 106/200
77/77 - 1s - loss: 0.2405 - accuracy: 0.9228 - val_loss: 0.8439 - val_accuracy: 0.8084 - 1s/epoch - 15ms/step
Epoch 107/200
77/77 - 1s - loss: 0.2205 - accuracy: 0.9233 - val_loss: 0.5514 - val_accuracy: 0.8411 - 1s/epoch - 15ms/step
Epoch 108/200
77/77 - 1s - loss: 0.2541 - accuracy: 0.9192 - val_loss: 0.4373 - val_accuracy: 0.8738 - 1s/epoch - 15ms/step
Epoch 109/200
77/77 - 1s - loss: 0.2216 - accuracy: 0.9181 - val_loss: 1.0460 - val_accuracy: 0.7220 - 1s/epoch - 15ms/step
Epoch 110/200
77/77 - 1s - loss: 0.2313 - accuracy: 0.9171 - val_loss: 1.4558 - val_accuracy: 0.6986 - 1s/epoch - 15ms/step
Epoch 111/200
77/77 - 1s - loss: 0.2573 - accuracy: 0.9114 - val_loss: 2.7137 - val_accuracy: 0.5748 - 1s/epoch - 15ms/step
Epoch 112/200
77/77 - 1s - loss: 0.2640 - accuracy: 0.9116 - val_loss: 0.8822 - val_accuracy: 0.7757 - 1s/epoch - 15ms/step
Epoch 113/200
77/77 - 1s - loss: 0.2680 - accuracy: 0.9108 - val_loss: 0.4436 - val_accuracy: 0.9042 - 1s/epoch - 15ms/step
Epoch 114/200
77/77 - 1s - loss: 0.2383 - accuracy: 0.9158 - val_loss: 1.7173 - val_accuracy: 0.6495 - 1s/epoch - 15ms/step
Epoch 115/200
77/77 - 1s - loss: 0.2203 - accuracy: 0.9225 - val_loss: 0.6289 - val_accuracy: 0.8294 - 1s/epoch - 15ms/step
Epoch 116/200
77/77 - 1s - loss: 0.2189 - accuracy: 0.9272 - val_loss: 0.4671 - val_accuracy: 0.8785 - 1s/epoch - 15ms/step
Epoch 117/200
77/77 - 1s - loss: 0.2645 - accuracy: 0.9108 - val_loss: 0.9991 - val_accuracy: 0.7453 - 1s/epoch - 15ms/step
Epoch 118/200
77/77 - 1s - loss: 0.2070 - accuracy: 0.9264 - val_loss: 0.7728 - val_accuracy: 0.8178 - 1s/epoch - 15ms/step
Epoch 119/200
77/77 - 1s - loss: 0.2480 - accuracy: 0.9223 - val_loss: 0.6796 - val_accuracy: 0.8248 - 1s/epoch - 15ms/step
Epoch 120/200
77/77 - 1s - loss: 0.2467 - accuracy: 0.9153 - val_loss: 1.0121 - val_accuracy: 0.7687 - 1s/epoch - 15ms/step
Epoch 121/200
77/77 - 1s - loss: 0.2795 - accuracy: 0.9010 - val_loss: 0.7508 - val_accuracy: 0.8131 - 1s/epoch - 15ms/step
Epoch 122/200
77/77 - 1s - loss: 0.2205 - accuracy: 0.9254 - val_loss: 0.6970 - val_accuracy: 0.8061 - 1s/epoch - 15ms/step
Epoch 123/200
77/77 - 1s - loss: 0.2321 - accuracy: 0.9272 - val_loss: 0.6723 - val_accuracy: 0.7944 - 1s/epoch - 15ms/step
Epoch 124/200
77/77 - 1s - loss: 0.2293 - accuracy: 0.9272 - val_loss: 0.4515 - val_accuracy: 0.8528 - 1s/epoch - 15ms/step
Epoch 125/200
77/77 - 1s - loss: 0.2092 - accuracy: 0.9277 - val_loss: 0.5681 - val_accuracy: 0.8341 - 1s/epoch - 15ms/step
Epoch 126/200
77/77 - 1s - loss: 0.2009 - accuracy: 0.9314 - val_loss: 0.9973 - val_accuracy: 0.7407 - 1s/epoch - 15ms/step
Epoch 127/200
77/77 - 1s - loss: 0.2339 - accuracy: 0.9272 - val_loss: 1.4975 - val_accuracy: 0.6565 - 1s/epoch - 15ms/step
Epoch 128/200
77/77 - 1s - loss: 0.1949 - accuracy: 0.9306 - val_loss: 0.7018 - val_accuracy: 0.8294 - 1s/epoch - 15ms/step
Epoch 129/200
77/77 - 1s - loss: 0.2036 - accuracy: 0.9363 - val_loss: 0.8387 - val_accuracy: 0.8037 - 1s/epoch - 15ms/step
Epoch 130/200
77/77 - 1s - loss: 0.1865 - accuracy: 0.9407 - val_loss: 0.9757 - val_accuracy: 0.7290 - 1s/epoch - 15ms/step
Epoch 131/200
77/77 - 1s - loss: 0.1874 - accuracy: 0.9379 - val_loss: 0.8847 - val_accuracy: 0.7804 - 1s/epoch - 15ms/step
Epoch 132/200
77/77 - 1s - loss: 0.1997 - accuracy: 0.9335 - val_loss: 0.9550 - val_accuracy: 0.7734 - 1s/epoch - 15ms/step
Epoch 133/200
77/77 - 1s - loss: 0.2434 - accuracy: 0.9270 - val_loss: 0.8541 - val_accuracy: 0.7617 - 1s/epoch - 15ms/step
Epoch 134/200
77/77 - 1s - loss: 0.2299 - accuracy: 0.9301 - val_loss: 1.7481 - val_accuracy: 0.6519 - 1s/epoch - 15ms/step
Epoch 135/200
77/77 - 1s - loss: 0.1749 - accuracy: 0.9441 - val_loss: 0.9120 - val_accuracy: 0.7780 - 1s/epoch - 15ms/step
Epoch 136/200
77/77 - 1s - loss: 0.2010 - accuracy: 0.9337 - val_loss: 0.5975 - val_accuracy: 0.8294 - 1s/epoch - 15ms/step
Epoch 137/200
77/77 - 1s - loss: 0.1928 - accuracy: 0.9358 - val_loss: 0.6672 - val_accuracy: 0.8107 - 1s/epoch - 15ms/step
Epoch 138/200
77/77 - 1s - loss: 0.1978 - accuracy: 0.9353 - val_loss: 0.5261 - val_accuracy: 0.8505 - 1s/epoch - 15ms/step
Epoch 139/200
77/77 - 1s - loss: 0.1645 - accuracy: 0.9480 - val_loss: 0.3613 - val_accuracy: 0.9042 - 1s/epoch - 15ms/step
Epoch 140/200
77/77 - 1s - loss: 0.2189 - accuracy: 0.9345 - val_loss: 0.9176 - val_accuracy: 0.7664 - 1s/epoch - 15ms/step
Epoch 141/200
77/77 - 1s - loss: 0.1797 - accuracy: 0.9452 - val_loss: 1.0065 - val_accuracy: 0.7430 - 1s/epoch - 15ms/step
Epoch 142/200
77/77 - 1s - loss: 0.2018 - accuracy: 0.9363 - val_loss: 1.3790 - val_accuracy: 0.7009 - 1s/epoch - 15ms/step
Epoch 143/200
77/77 - 1s - loss: 0.1925 - accuracy: 0.9348 - val_loss: 1.1297 - val_accuracy: 0.7336 - 1s/epoch - 15ms/step
Epoch 144/200
77/77 - 1s - loss: 0.2022 - accuracy: 0.9358 - val_loss: 1.4766 - val_accuracy: 0.6893 - 1s/epoch - 15ms/step
Epoch 145/200
77/77 - 1s - loss: 0.1750 - accuracy: 0.9485 - val_loss: 0.8584 - val_accuracy: 0.8037 - 1s/epoch - 15ms/step
Epoch 146/200
77/77 - 1s - loss: 0.1716 - accuracy: 0.9407 - val_loss: 2.0207 - val_accuracy: 0.6028 - 1s/epoch - 15ms/step
Epoch 147/200
77/77 - 1s - loss: 0.1765 - accuracy: 0.9446 - val_loss: 0.8974 - val_accuracy: 0.7921 - 1s/epoch - 15ms/step
Epoch 148/200
77/77 - 1s - loss: 0.2202 - accuracy: 0.9293 - val_loss: 1.2410 - val_accuracy: 0.6846 - 1s/epoch - 15ms/step
Epoch 149/200
77/77 - 1s - loss: 0.1612 - accuracy: 0.9452 - val_loss: 1.3236 - val_accuracy: 0.7243 - 1s/epoch - 15ms/step
Epoch 150/200
77/77 - 1s - loss: 0.1870 - accuracy: 0.9400 - val_loss: 0.7305 - val_accuracy: 0.8248 - 1s/epoch - 15ms/step
Epoch 151/200
77/77 - 1s - loss: 0.1559 - accuracy: 0.9524 - val_loss: 0.4713 - val_accuracy: 0.8832 - 1s/epoch - 15ms/step
Epoch 152/200
77/77 - 1s - loss: 0.1499 - accuracy: 0.9537 - val_loss: 0.4667 - val_accuracy: 0.8785 - 1s/epoch - 15ms/step
Epoch 153/200
77/77 - 1s - loss: 0.1912 - accuracy: 0.9350 - val_loss: 0.9394 - val_accuracy: 0.7593 - 1s/epoch - 15ms/step
Epoch 154/200
77/77 - 1s - loss: 0.1861 - accuracy: 0.9415 - val_loss: 4.0100 - val_accuracy: 0.4322 - 1s/epoch - 15ms/step
Epoch 155/200
77/77 - 1s - loss: 0.2066 - accuracy: 0.9340 - val_loss: 0.7759 - val_accuracy: 0.8154 - 1s/epoch - 15ms/step
Epoch 156/200
77/77 - 1s - loss: 0.2022 - accuracy: 0.9324 - val_loss: 0.7389 - val_accuracy: 0.8107 - 1s/epoch - 15ms/step
Epoch 157/200
77/77 - 1s - loss: 0.1559 - accuracy: 0.9496 - val_loss: 0.4767 - val_accuracy: 0.8738 - 1s/epoch - 15ms/step
Epoch 158/200
77/77 - 1s - loss: 0.1561 - accuracy: 0.9504 - val_loss: 0.6723 - val_accuracy: 0.8341 - 1s/epoch - 15ms/step
Epoch 159/200
77/77 - 1s - loss: 0.1450 - accuracy: 0.9540 - val_loss: 1.8042 - val_accuracy: 0.6986 - 1s/epoch - 15ms/step
Epoch 160/200
77/77 - 1s - loss: 0.1708 - accuracy: 0.9454 - val_loss: 1.3233 - val_accuracy: 0.7126 - 1s/epoch - 15ms/step
Epoch 161/200
77/77 - 1s - loss: 0.1556 - accuracy: 0.9480 - val_loss: 1.2577 - val_accuracy: 0.7360 - 1s/epoch - 18ms/step
Epoch 162/200
77/77 - 1s - loss: 0.1678 - accuracy: 0.9480 - val_loss: 1.1420 - val_accuracy: 0.7710 - 1s/epoch - 19ms/step
Epoch 163/200
77/77 - 1s - loss: 0.1861 - accuracy: 0.9342 - val_loss: 0.7558 - val_accuracy: 0.8411 - 1s/epoch - 18ms/step
Epoch 164/200
77/77 - 1s - loss: 0.1737 - accuracy: 0.9436 - val_loss: 1.5686 - val_accuracy: 0.6589 - 1s/epoch - 15ms/step
Epoch 165/200
77/77 - 1s - loss: 0.1800 - accuracy: 0.9444 - val_loss: 1.0370 - val_accuracy: 0.7710 - 1s/epoch - 15ms/step
Epoch 166/200
77/77 - 1s - loss: 0.1503 - accuracy: 0.9543 - val_loss: 1.2218 - val_accuracy: 0.7593 - 1s/epoch - 15ms/step
Epoch 167/200
77/77 - 1s - loss: 0.1462 - accuracy: 0.9558 - val_loss: 0.8459 - val_accuracy: 0.8131 - 1s/epoch - 15ms/step
Epoch 168/200
77/77 - 1s - loss: 0.1229 - accuracy: 0.9581 - val_loss: 0.4940 - val_accuracy: 0.8925 - 1s/epoch - 15ms/step
Epoch 169/200
77/77 - 1s - loss: 0.1561 - accuracy: 0.9511 - val_loss: 0.5388 - val_accuracy: 0.8715 - 1s/epoch - 15ms/step
Epoch 170/200
77/77 - 1s - loss: 0.1536 - accuracy: 0.9480 - val_loss: 0.6893 - val_accuracy: 0.8505 - 1s/epoch - 15ms/step
Epoch 171/200
77/77 - 1s - loss: 0.1631 - accuracy: 0.9496 - val_loss: 0.5194 - val_accuracy: 0.8855 - 1s/epoch - 15ms/step
Epoch 172/200
77/77 - 1s - loss: 0.1403 - accuracy: 0.9571 - val_loss: 3.1395 - val_accuracy: 0.5981 - 1s/epoch - 15ms/step
Epoch 173/200
77/77 - 1s - loss: 0.1439 - accuracy: 0.9509 - val_loss: 0.4587 - val_accuracy: 0.8785 - 1s/epoch - 16ms/step
Epoch 174/200
77/77 - 1s - loss: 0.1405 - accuracy: 0.9558 - val_loss: 0.4486 - val_accuracy: 0.8879 - 1s/epoch - 15ms/step
Epoch 175/200
77/77 - 1s - loss: 0.1526 - accuracy: 0.9511 - val_loss: 1.6330 - val_accuracy: 0.6682 - 1s/epoch - 15ms/step
Epoch 176/200
77/77 - 1s - loss: 0.1133 - accuracy: 0.9602 - val_loss: 1.0183 - val_accuracy: 0.7944 - 1s/epoch - 15ms/step
Epoch 177/200
77/77 - 1s - loss: 0.1381 - accuracy: 0.9581 - val_loss: 1.0041 - val_accuracy: 0.7850 - 1s/epoch - 15ms/step
Epoch 178/200
77/77 - 1s - loss: 0.1627 - accuracy: 0.9485 - val_loss: 1.7937 - val_accuracy: 0.6682 - 1s/epoch - 15ms/step
Epoch 179/200
77/77 - 1s - loss: 0.1382 - accuracy: 0.9555 - val_loss: 1.2323 - val_accuracy: 0.7196 - 1s/epoch - 15ms/step
Epoch 180/200
77/77 - 1s - loss: 0.1669 - accuracy: 0.9524 - val_loss: 0.9157 - val_accuracy: 0.7874 - 1s/epoch - 15ms/step
Epoch 181/200
77/77 - 1s - loss: 0.1320 - accuracy: 0.9592 - val_loss: 0.5499 - val_accuracy: 0.8715 - 1s/epoch - 15ms/step
Epoch 182/200
77/77 - 1s - loss: 0.1430 - accuracy: 0.9600 - val_loss: 0.5914 - val_accuracy: 0.8528 - 1s/epoch - 15ms/step
Epoch 183/200
77/77 - 1s - loss: 0.1550 - accuracy: 0.9511 - val_loss: 0.6271 - val_accuracy: 0.8528 - 1s/epoch - 15ms/step
Epoch 184/200
77/77 - 1s - loss: 0.1230 - accuracy: 0.9636 - val_loss: 0.4785 - val_accuracy: 0.8949 - 1s/epoch - 15ms/step
Epoch 185/200
77/77 - 1s - loss: 0.1955 - accuracy: 0.9379 - val_loss: 0.8892 - val_accuracy: 0.7967 - 1s/epoch - 15ms/step
Epoch 186/200
77/77 - 1s - loss: 0.1287 - accuracy: 0.9581 - val_loss: 0.4366 - val_accuracy: 0.8902 - 1s/epoch - 15ms/step
Epoch 187/200
77/77 - 1s - loss: 0.1219 - accuracy: 0.9633 - val_loss: 0.5484 - val_accuracy: 0.8715 - 1s/epoch - 15ms/step
Epoch 188/200
77/77 - 1s - loss: 0.1277 - accuracy: 0.9594 - val_loss: 0.5518 - val_accuracy: 0.8621 - 1s/epoch - 15ms/step
Epoch 189/200
77/77 - 1s - loss: 0.1971 - accuracy: 0.9472 - val_loss: 0.8884 - val_accuracy: 0.8014 - 1s/epoch - 15ms/step
Epoch 190/200
77/77 - 1s - loss: 0.1691 - accuracy: 0.9415 - val_loss: 0.3845 - val_accuracy: 0.9065 - 1s/epoch - 15ms/step
Epoch 191/200
77/77 - 1s - loss: 0.1375 - accuracy: 0.9592 - val_loss: 0.4315 - val_accuracy: 0.8738 - 1s/epoch - 15ms/step
Epoch 192/200
77/77 - 1s - loss: 0.1400 - accuracy: 0.9553 - val_loss: 0.4341 - val_accuracy: 0.8808 - 1s/epoch - 15ms/step
Epoch 193/200
77/77 - 1s - loss: 0.1329 - accuracy: 0.9620 - val_loss: 1.2771 - val_accuracy: 0.7150 - 1s/epoch - 15ms/step
Epoch 194/200
77/77 - 1s - loss: 0.1350 - accuracy: 0.9594 - val_loss: 0.9742 - val_accuracy: 0.7664 - 1s/epoch - 15ms/step
Epoch 195/200
77/77 - 1s - loss: 0.0990 - accuracy: 0.9683 - val_loss: 0.4587 - val_accuracy: 0.8902 - 1s/epoch - 15ms/step
Epoch 196/200
77/77 - 1s - loss: 0.1321 - accuracy: 0.9623 - val_loss: 1.0273 - val_accuracy: 0.7664 - 1s/epoch - 15ms/step
Epoch 197/200
77/77 - 1s - loss: 0.1665 - accuracy: 0.9475 - val_loss: 2.4706 - val_accuracy: 0.5070 - 1s/epoch - 15ms/step
Epoch 198/200
77/77 - 1s - loss: 0.1481 - accuracy: 0.9610 - val_loss: 0.7776 - val_accuracy: 0.8107 - 1s/epoch - 15ms/step
Epoch 199/200
77/77 - 1s - loss: 0.1552 - accuracy: 0.9540 - val_loss: 0.4528 - val_accuracy: 0.8879 - 1s/epoch - 15ms/step
Epoch 200/200
77/77 - 1s - loss: 0.1050 - accuracy: 0.9675 - val_loss: 0.4881 - val_accuracy: 0.8808 - 1s/epoch - 15ms/step

Model Evaluation

plt.plot(history_1.history['accuracy'])
plt.plot(history_1.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()

Observations:

• We can see from the above plot that the training accuracy of the model and the validation accuracy are good but there is variance that can be tuned further.
• The model seems to be having a varying overfit on the data
• We can check on further tuning on the models to achieve a higher accuracy

Model Evaluation on Training dataset

model.evaluate(X_train_normalized, y_train_encoded, verbose=1)
model1_TrainDF = model_performance_classification_sklearn_with_threshold(
    model, X_train_normalized, y_train_encoded,
)
model1_TrainDF

121/121 [==============================] - 1s 7ms/step - loss: 0.0782 - accuracy: 0.9737
121/121 [==============================] - 1s 5ms/step

	Accuracy	Recall	Precision	F1
0	0.973746	0.973746	0.974874	0.973941

Observations

• The Training accuracy of this model is 97.4% and the other metrics also are in the same range. It has a good accuracy score
• We can observe that some of the classes are not predicted correctly
• Except for the classes 1, 2, 5, 9 & 10 which are well classified, the other classes are not classified properly with some false classifications

Model Evaluation on Validation dataset

model.evaluate(X_val_normalized, y_val_encoded, verbose=1)
model1_ValDF = model_performance_classification_sklearn_with_threshold(
    model, X_val_normalized,y_val_encoded,
)
model1_ValDF

14/14 [==============================] - 0s 15ms/step - loss: 0.4881 - accuracy: 0.8808
14/14 [==============================] - 0s 6ms/step

	Accuracy	Recall	Precision	F1
0	0.880841	0.880841	0.887867	0.881465

Observations

• The Validation accuracy of this model is 88% and the other metrics also are in the same range. Precision has a higher metric
• When compared with the Training accurace, there is ~10% variance in the Accuracy scores. There is room for improvement to reduce this variance

Model Evaluation on Test dataset

model.evaluate(X_test_normalized, y_test_encoded, verbose=2)
model1_TestDF = model_performance_classification_sklearn_with_threshold(model, X_test_normalized, y_test_encoded)
model1_TestDF

15/15 - 0s - loss: 0.5750 - accuracy: 0.8863 - 207ms/epoch - 14ms/step
15/15 [==============================] - 0s 5ms/step

	Accuracy	Recall	Precision	F1
0	0.886316	0.886316	0.894756	0.887596

Observations:

• The Testing accuracy of this model is 88.6% and the other metrics also are in the same range. The accuracy is very good and also the Precision has a higher metric which is good. This would help to determine more positive predictions
• When compared with the Training accuracy (97.4%), the Testing accuracy looks very good with only around 9% variance between the two
• Using the classification chart, we can observe that the classes are classified correctly for most of the classes except for the classes 0, 6, 8 & 9. We can look at tuning it further to achieve better accuracy

Generating the predictions using test data

y_pred=model.predict(X_test_normalized)
y_pred

15/15 [==============================] - 0s 5ms/step

array([[2.5825295e-10, 1.8731525e-11, 1.0227022e-10, ..., 7.6328673e-05,
        3.5734458e-05, 6.1272149e-06],
       [3.6871587e-08, 1.3989663e-03, 1.0832313e-03, ..., 2.1792054e-03,
        9.9502015e-01, 3.6656304e-05],
       [4.4761779e-07, 6.3672098e-03, 4.6234634e-03, ..., 6.0020573e-03,
        9.8069340e-01, 3.5219963e-04],
       ...,
       [9.8092252e-01, 8.8385230e-11, 1.7658030e-08, ..., 2.3761793e-12,
        5.3023858e-10, 3.4472514e-09],
       [6.8217845e-09, 3.8922789e-12, 3.1978364e-09, ..., 5.4159387e-05,
        1.3753737e-11, 1.8412022e-06],
       [1.6231765e-03, 2.2416206e-02, 1.8579045e-01, ..., 7.3115951e-03,
        1.3829044e-01, 6.7066081e-02]], dtype=float32)

# Visualizing the predicted and correct label of images from test data
rows = 1                                                                        
cols = 5                                                                        
for i in range(cols):
    for j in range(rows):
      random_index = np.random.randint(0, 400)    
      plt.imshow(X_test[random_index])
      plt.show()

      print('Predicted Label', enc.inverse_transform(model.predict((X_test_normalized[random_index].reshape(1,64,64,3)))))   
      print('True Label', enc.inverse_transform(y_test_encoded)[random_index])       

1/1 [==============================] - 0s 85ms/step
Predicted Label ['Sugar beet']
True Label Sugar beet

1/1 [==============================] - 0s 17ms/step
Predicted Label ['Black-grass']
True Label Loose Silky-bent

1/1 [==============================] - 0s 17ms/step
Predicted Label ['Loose Silky-bent']
True Label Loose Silky-bent

1/1 [==============================] - 0s 18ms/step
Predicted Label ['Scentless Mayweed']
True Label Scentless Mayweed

1/1 [==============================] - 0s 20ms/step
Predicted Label ['Cleavers']
True Label Small-flowered Cranesbill

Inference:

• We can observe that some of the random images has few incorrect classifications in this model.

Model 2 - CNN with Data Augumentation

---

Creating a CNN model sequentially by using Data Augumentation to check if we can improve the model performance and metrics further

# Clearing backend
backend.clear_session()

# Fixing the seed for random number generators
import random
np.random.seed(42)
random.seed(42)
tf.random.set_seed(42)

# All images to be rescaled by 1/255.
# Applying the Data Augmentation levers
train_datagen = ImageDataGenerator(
                              horizontal_flip = True,
                              vertical_flip = True, 
                              height_shift_range= 0.1, 
                              width_shift_range=0.1, 
                              rotation_range=20, 
                              shear_range = 0.1,
                              zoom_range=0.1, 
                              fill_mode='nearest'
                              )
#test_datagen  = ImageDataGenerator(rescale = 1.0/255.)

# Intializing a sequential model
model = Sequential()

# Adding first conv layer with 64 filters and kernel size 3x3 , padding 'same' provides the output size same as the input size
# Input_shape denotes input image dimension of images
# Adding first conv layer with 64 filters and kernel size 3x3 , padding 'same' provides the output size same as the input size
# Input_shape denotes input image dimension of images
model.add(Conv2D(64, (3, 3), activation='relu', padding="same", input_shape=(64, 64, 3)))
# Adding max pooling to reduce the size of output of first conv layer
model.add(MaxPooling2D(2, 2))
model.add(BatchNormalization())

model.add(Conv2D(64, (3, 3), activation='relu', padding="same"))
model.add(BatchNormalization())
model.add(MaxPooling2D(2, 2))
model.add(Dense(32, activation='relu'))

model.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model.add(MaxPooling2D(2, 2))
model.add(Dropout(0.3))

model.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model.add(Dense(32, activation='relu'))
model.add(MaxPooling2D(2, 2))
model.add(Dropout(0.3))

model.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model.add(MaxPooling2D(2, 2))

model.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model.add(MaxPooling2D(2, 2))
model.add(BatchNormalization())

# flattening the output of the conv layer after max pooling to make it ready for creating dense connections
model.add(Flatten())

# Adding a fully connected dense layer with 100 neurons    
model.add(Dense(64, activation='relu'))
model.add(Dropout(0.3))

# Adding a fully connected dense layer with 100 neurons    
model.add(Dense(32, activation='relu'))
model.add(Dropout(0.3))

model.add(Dense(32, activation='relu'))

# Adding the output layer with 12 neurons and activation functions as softmax since this is a multi-class classification problem  
model.add(Dense(12, activation='softmax'))

# Using Adam Optimizer
opt=Adam()

# Compile model
model.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])

# Generating the summary of the model
model.summary()

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d (Conv2D)             (None, 64, 64, 64)        1792      
                                                                 
 max_pooling2d (MaxPooling2D  (None, 32, 32, 64)       0         
 )                                                               
                                                                 
 batch_normalization (BatchN  (None, 32, 32, 64)       256       
 ormalization)                                                   
                                                                 
 conv2d_1 (Conv2D)           (None, 32, 32, 64)        36928     
                                                                 
 batch_normalization_1 (Batc  (None, 32, 32, 64)       256       
 hNormalization)                                                 
                                                                 
 max_pooling2d_1 (MaxPooling  (None, 16, 16, 64)       0         
 2D)                                                             
                                                                 
 dense (Dense)               (None, 16, 16, 32)        2080      
                                                                 
 conv2d_2 (Conv2D)           (None, 16, 16, 32)        9248      
                                                                 
 max_pooling2d_2 (MaxPooling  (None, 8, 8, 32)         0         
 2D)                                                             
                                                                 
 dropout (Dropout)           (None, 8, 8, 32)          0         
                                                                 
 conv2d_3 (Conv2D)           (None, 8, 8, 32)          9248      
                                                                 
 dense_1 (Dense)             (None, 8, 8, 32)          1056      
                                                                 
 max_pooling2d_3 (MaxPooling  (None, 4, 4, 32)         0         
 2D)                                                             
                                                                 
 dropout_1 (Dropout)         (None, 4, 4, 32)          0         
                                                                 
 conv2d_4 (Conv2D)           (None, 4, 4, 32)          9248      
                                                                 
 max_pooling2d_4 (MaxPooling  (None, 2, 2, 32)         0         
 2D)                                                             
                                                                 
 conv2d_5 (Conv2D)           (None, 2, 2, 32)          9248      
                                                                 
 max_pooling2d_5 (MaxPooling  (None, 1, 1, 32)         0         
 2D)                                                             
                                                                 
 batch_normalization_2 (Batc  (None, 1, 1, 32)         128       
 hNormalization)                                                 
                                                                 
 flatten (Flatten)           (None, 32)                0         
                                                                 
 dense_2 (Dense)             (None, 64)                2112      
                                                                 
 dropout_2 (Dropout)         (None, 64)                0         
                                                                 
 dense_3 (Dense)             (None, 32)                2080      
                                                                 
 dropout_3 (Dropout)         (None, 32)                0         
                                                                 
 dense_4 (Dense)             (None, 32)                1056      
                                                                 
 dense_5 (Dense)             (None, 12)                396       
                                                                 
=================================================================
Total params: 85,132
Trainable params: 84,812
Non-trainable params: 320
_________________________________________________________________

Observations:

• There are a total of 85,132 params with 84,812 trainable params
• We have applied Convolution along with BatchNormalization, pooling and droputs to tune the model along with the flat layer
• We will appy Data Augmentation on the Imgae dataset by applying the various parameters
• We are using the loss function - categorical_crossentropy as this is a multi-class classification problem and Adam as the optimizer
• Also, we are choosing accuracy as the metric to measure the performance of the model

Fitting the model on the train data

• Applying Data Augumentation function on the training images along with Epochs & Batch Size parameters.

# Epochs
epochs = 200
# Batch size
batch_size = 30

history = model.fit(train_datagen.flow(X_train_normalized,y_train_encoded,
                                       batch_size=batch_size, 
                                       seed=42,
                                       shuffle=False),
                    epochs=epochs,
                    steps_per_epoch=X_train_normalized.shape[0] // batch_size,
                    validation_data=(X_val_normalized,y_val_encoded),
                    verbose=1)

Epoch 1/200
128/128 [==============================] - 7s 44ms/step - loss: 2.4939 - accuracy: 0.1124 - val_loss: 2.4278 - val_accuracy: 0.1028
Epoch 2/200
128/128 [==============================] - 5s 41ms/step - loss: 2.3580 - accuracy: 0.1763 - val_loss: 2.4823 - val_accuracy: 0.1285
Epoch 3/200
128/128 [==============================] - 5s 41ms/step - loss: 1.9786 - accuracy: 0.3212 - val_loss: 2.4924 - val_accuracy: 0.1682
Epoch 4/200
128/128 [==============================] - 5s 42ms/step - loss: 1.7392 - accuracy: 0.3752 - val_loss: 4.4700 - val_accuracy: 0.1659
Epoch 5/200
128/128 [==============================] - 5s 40ms/step - loss: 1.6527 - accuracy: 0.4192 - val_loss: 1.9388 - val_accuracy: 0.3318
Epoch 6/200
128/128 [==============================] - 5s 41ms/step - loss: 1.5601 - accuracy: 0.4396 - val_loss: 2.0940 - val_accuracy: 0.2103
Epoch 7/200
128/128 [==============================] - 5s 41ms/step - loss: 1.5004 - accuracy: 0.4600 - val_loss: 1.2419 - val_accuracy: 0.5654
Epoch 8/200
128/128 [==============================] - 5s 41ms/step - loss: 1.4386 - accuracy: 0.4894 - val_loss: 3.1199 - val_accuracy: 0.2570
Epoch 9/200
128/128 [==============================] - 5s 41ms/step - loss: 1.3806 - accuracy: 0.5185 - val_loss: 1.4433 - val_accuracy: 0.5187
Epoch 10/200
128/128 [==============================] - 5s 42ms/step - loss: 1.3531 - accuracy: 0.5342 - val_loss: 1.3506 - val_accuracy: 0.5327
Epoch 11/200
128/128 [==============================] - 5s 41ms/step - loss: 1.3271 - accuracy: 0.5345 - val_loss: 3.3043 - val_accuracy: 0.2921
Epoch 12/200
128/128 [==============================] - 5s 41ms/step - loss: 1.2743 - accuracy: 0.5507 - val_loss: 1.7082 - val_accuracy: 0.4182
Epoch 13/200
128/128 [==============================] - 5s 40ms/step - loss: 1.2477 - accuracy: 0.5638 - val_loss: 3.0424 - val_accuracy: 0.2477
Epoch 14/200
128/128 [==============================] - 5s 41ms/step - loss: 1.2089 - accuracy: 0.5858 - val_loss: 1.4220 - val_accuracy: 0.4883
Epoch 15/200
128/128 [==============================] - 5s 40ms/step - loss: 1.1833 - accuracy: 0.5910 - val_loss: 1.2917 - val_accuracy: 0.5514
Epoch 16/200
128/128 [==============================] - 5s 41ms/step - loss: 1.1294 - accuracy: 0.6204 - val_loss: 2.0679 - val_accuracy: 0.3715
Epoch 17/200
128/128 [==============================] - 5s 42ms/step - loss: 1.1005 - accuracy: 0.6246 - val_loss: 1.0766 - val_accuracy: 0.6098
Epoch 18/200
128/128 [==============================] - 5s 41ms/step - loss: 1.0635 - accuracy: 0.6440 - val_loss: 2.3211 - val_accuracy: 0.2991
Epoch 19/200
128/128 [==============================] - 5s 41ms/step - loss: 1.0392 - accuracy: 0.6531 - val_loss: 2.1380 - val_accuracy: 0.3832
Epoch 20/200
128/128 [==============================] - 5s 41ms/step - loss: 1.0123 - accuracy: 0.6474 - val_loss: 1.8464 - val_accuracy: 0.4322
Epoch 21/200
128/128 [==============================] - 5s 41ms/step - loss: 0.9680 - accuracy: 0.6709 - val_loss: 1.2289 - val_accuracy: 0.5724
Epoch 22/200
128/128 [==============================] - 5s 41ms/step - loss: 0.9880 - accuracy: 0.6744 - val_loss: 1.2680 - val_accuracy: 0.5444
Epoch 23/200
128/128 [==============================] - 6s 44ms/step - loss: 0.9614 - accuracy: 0.6796 - val_loss: 0.9850 - val_accuracy: 0.6986
Epoch 24/200
128/128 [==============================] - 5s 41ms/step - loss: 0.9182 - accuracy: 0.6838 - val_loss: 1.0132 - val_accuracy: 0.6589
Epoch 25/200
128/128 [==============================] - 6s 44ms/step - loss: 0.8811 - accuracy: 0.7089 - val_loss: 2.7736 - val_accuracy: 0.3762
Epoch 26/200
128/128 [==============================] - 5s 41ms/step - loss: 0.8731 - accuracy: 0.7092 - val_loss: 0.9394 - val_accuracy: 0.6939
Epoch 27/200
128/128 [==============================] - 5s 41ms/step - loss: 0.8318 - accuracy: 0.7247 - val_loss: 1.6999 - val_accuracy: 0.5140
Epoch 28/200
128/128 [==============================] - 5s 41ms/step - loss: 0.8553 - accuracy: 0.7226 - val_loss: 0.7422 - val_accuracy: 0.7500
Epoch 29/200
128/128 [==============================] - 5s 41ms/step - loss: 0.8079 - accuracy: 0.7357 - val_loss: 1.0259 - val_accuracy: 0.6355
Epoch 30/200
128/128 [==============================] - 5s 41ms/step - loss: 0.8222 - accuracy: 0.7307 - val_loss: 2.4390 - val_accuracy: 0.2290
Epoch 31/200
128/128 [==============================] - 5s 41ms/step - loss: 0.7719 - accuracy: 0.7364 - val_loss: 0.9686 - val_accuracy: 0.6939
Epoch 32/200
128/128 [==============================] - 5s 42ms/step - loss: 0.7861 - accuracy: 0.7404 - val_loss: 0.8620 - val_accuracy: 0.7056
Epoch 33/200
128/128 [==============================] - 5s 41ms/step - loss: 0.7826 - accuracy: 0.7383 - val_loss: 1.0813 - val_accuracy: 0.6449
Epoch 34/200
128/128 [==============================] - 5s 41ms/step - loss: 0.7701 - accuracy: 0.7480 - val_loss: 0.9110 - val_accuracy: 0.6822
Epoch 35/200
128/128 [==============================] - 5s 40ms/step - loss: 0.7456 - accuracy: 0.7522 - val_loss: 1.5704 - val_accuracy: 0.6005
Epoch 36/200
128/128 [==============================] - 5s 42ms/step - loss: 0.7342 - accuracy: 0.7561 - val_loss: 4.0755 - val_accuracy: 0.2921
Epoch 37/200
128/128 [==============================] - 5s 41ms/step - loss: 0.7279 - accuracy: 0.7642 - val_loss: 0.9836 - val_accuracy: 0.6752
Epoch 38/200
128/128 [==============================] - 5s 41ms/step - loss: 0.7088 - accuracy: 0.7616 - val_loss: 1.7569 - val_accuracy: 0.4953
Epoch 39/200
128/128 [==============================] - 5s 41ms/step - loss: 0.7184 - accuracy: 0.7582 - val_loss: 1.4326 - val_accuracy: 0.5397
Epoch 40/200
128/128 [==============================] - 8s 63ms/step - loss: 0.6835 - accuracy: 0.7739 - val_loss: 0.9249 - val_accuracy: 0.6752
Epoch 41/200
128/128 [==============================] - 5s 42ms/step - loss: 0.6607 - accuracy: 0.7794 - val_loss: 2.1595 - val_accuracy: 0.4743
Epoch 42/200
128/128 [==============================] - 6s 43ms/step - loss: 0.6860 - accuracy: 0.7681 - val_loss: 1.0651 - val_accuracy: 0.6565
Epoch 43/200
128/128 [==============================] - 5s 41ms/step - loss: 0.6546 - accuracy: 0.7818 - val_loss: 0.7759 - val_accuracy: 0.7313
Epoch 44/200
128/128 [==============================] - 5s 41ms/step - loss: 0.6592 - accuracy: 0.7755 - val_loss: 1.1068 - val_accuracy: 0.6379
Epoch 45/200
128/128 [==============================] - 6s 44ms/step - loss: 0.6410 - accuracy: 0.7799 - val_loss: 1.1661 - val_accuracy: 0.6752
Epoch 46/200
128/128 [==============================] - 5s 41ms/step - loss: 0.6189 - accuracy: 0.7891 - val_loss: 0.9643 - val_accuracy: 0.7033
Epoch 47/200
128/128 [==============================] - 5s 41ms/step - loss: 0.6407 - accuracy: 0.7750 - val_loss: 0.5878 - val_accuracy: 0.8084
Epoch 48/200
128/128 [==============================] - 5s 40ms/step - loss: 0.6360 - accuracy: 0.7883 - val_loss: 0.7406 - val_accuracy: 0.7430
Epoch 49/200
128/128 [==============================] - 5s 42ms/step - loss: 0.5796 - accuracy: 0.7951 - val_loss: 0.7106 - val_accuracy: 0.7710
Epoch 50/200
128/128 [==============================] - 5s 41ms/step - loss: 0.6240 - accuracy: 0.7915 - val_loss: 0.7714 - val_accuracy: 0.7640
Epoch 51/200
128/128 [==============================] - 5s 41ms/step - loss: 0.6180 - accuracy: 0.7901 - val_loss: 0.6777 - val_accuracy: 0.7710
Epoch 52/200
128/128 [==============================] - 5s 41ms/step - loss: 0.6130 - accuracy: 0.7975 - val_loss: 0.5645 - val_accuracy: 0.8131
Epoch 53/200
128/128 [==============================] - 5s 41ms/step - loss: 0.6128 - accuracy: 0.7899 - val_loss: 0.4666 - val_accuracy: 0.8458
Epoch 54/200
128/128 [==============================] - 5s 41ms/step - loss: 0.5919 - accuracy: 0.7957 - val_loss: 0.9996 - val_accuracy: 0.7173
Epoch 55/200
128/128 [==============================] - 5s 41ms/step - loss: 0.5692 - accuracy: 0.8046 - val_loss: 0.6480 - val_accuracy: 0.7804
Epoch 56/200
128/128 [==============================] - 7s 52ms/step - loss: 0.5662 - accuracy: 0.8080 - val_loss: 1.4497 - val_accuracy: 0.6308
Epoch 57/200
128/128 [==============================] - 6s 43ms/step - loss: 0.5657 - accuracy: 0.8108 - val_loss: 0.5125 - val_accuracy: 0.8154
Epoch 58/200
128/128 [==============================] - 5s 41ms/step - loss: 0.5619 - accuracy: 0.8067 - val_loss: 0.6239 - val_accuracy: 0.8061
Epoch 59/200
128/128 [==============================] - 5s 41ms/step - loss: 0.5319 - accuracy: 0.8184 - val_loss: 0.4419 - val_accuracy: 0.8505
Epoch 60/200
128/128 [==============================] - 5s 42ms/step - loss: 0.5332 - accuracy: 0.8221 - val_loss: 0.5705 - val_accuracy: 0.8178
Epoch 61/200
128/128 [==============================] - 5s 41ms/step - loss: 0.5996 - accuracy: 0.7943 - val_loss: 0.9098 - val_accuracy: 0.7523
Epoch 62/200
128/128 [==============================] - 5s 41ms/step - loss: 0.5390 - accuracy: 0.8190 - val_loss: 0.7491 - val_accuracy: 0.7477
Epoch 63/200
128/128 [==============================] - 5s 41ms/step - loss: 0.5348 - accuracy: 0.8213 - val_loss: 1.4596 - val_accuracy: 0.6379
Epoch 64/200
128/128 [==============================] - 5s 41ms/step - loss: 0.5295 - accuracy: 0.8213 - val_loss: 0.8024 - val_accuracy: 0.7640
Epoch 65/200
128/128 [==============================] - 6s 43ms/step - loss: 0.5021 - accuracy: 0.8313 - val_loss: 0.5257 - val_accuracy: 0.8294
Epoch 66/200
128/128 [==============================] - 5s 41ms/step - loss: 0.5327 - accuracy: 0.8258 - val_loss: 0.3970 - val_accuracy: 0.8692
Epoch 67/200
128/128 [==============================] - 5s 42ms/step - loss: 0.5031 - accuracy: 0.8229 - val_loss: 0.6878 - val_accuracy: 0.7617
Epoch 68/200
128/128 [==============================] - 5s 41ms/step - loss: 0.5057 - accuracy: 0.8276 - val_loss: 1.3814 - val_accuracy: 0.6192
Epoch 69/200
128/128 [==============================] - 5s 41ms/step - loss: 0.5169 - accuracy: 0.8266 - val_loss: 0.5553 - val_accuracy: 0.8084
Epoch 70/200
128/128 [==============================] - 5s 41ms/step - loss: 0.5019 - accuracy: 0.8305 - val_loss: 0.5101 - val_accuracy: 0.8224
Epoch 71/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4820 - accuracy: 0.8407 - val_loss: 1.7412 - val_accuracy: 0.6285
Epoch 72/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4772 - accuracy: 0.8329 - val_loss: 0.4152 - val_accuracy: 0.8598
Epoch 73/200
128/128 [==============================] - 6s 44ms/step - loss: 0.5052 - accuracy: 0.8279 - val_loss: 1.2373 - val_accuracy: 0.6168
Epoch 74/200
128/128 [==============================] - 5s 42ms/step - loss: 0.4703 - accuracy: 0.8480 - val_loss: 0.4226 - val_accuracy: 0.8645
Epoch 75/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4565 - accuracy: 0.8402 - val_loss: 0.6687 - val_accuracy: 0.7897
Epoch 76/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4403 - accuracy: 0.8480 - val_loss: 0.5239 - val_accuracy: 0.8271
Epoch 77/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4899 - accuracy: 0.8357 - val_loss: 1.4715 - val_accuracy: 0.6192
Epoch 78/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4830 - accuracy: 0.8394 - val_loss: 0.4089 - val_accuracy: 0.8505
Epoch 79/200
128/128 [==============================] - 5s 42ms/step - loss: 0.4525 - accuracy: 0.8475 - val_loss: 0.6508 - val_accuracy: 0.7944
Epoch 80/200
128/128 [==============================] - 6s 43ms/step - loss: 0.4094 - accuracy: 0.8619 - val_loss: 0.6761 - val_accuracy: 0.7991
Epoch 81/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4596 - accuracy: 0.8402 - val_loss: 1.0543 - val_accuracy: 0.6145
Epoch 82/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4643 - accuracy: 0.8334 - val_loss: 1.0630 - val_accuracy: 0.7243
Epoch 83/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4530 - accuracy: 0.8436 - val_loss: 0.4625 - val_accuracy: 0.8364
Epoch 84/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4458 - accuracy: 0.8541 - val_loss: 1.2329 - val_accuracy: 0.6542
Epoch 85/200
128/128 [==============================] - 6s 43ms/step - loss: 0.4176 - accuracy: 0.8564 - val_loss: 0.6211 - val_accuracy: 0.7874
Epoch 86/200
128/128 [==============================] - 6s 48ms/step - loss: 0.4693 - accuracy: 0.8444 - val_loss: 0.6672 - val_accuracy: 0.7967
Epoch 87/200
128/128 [==============================] - 6s 43ms/step - loss: 0.4379 - accuracy: 0.8549 - val_loss: 0.9556 - val_accuracy: 0.6916
Epoch 88/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4032 - accuracy: 0.8604 - val_loss: 0.4186 - val_accuracy: 0.8832
Epoch 89/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4087 - accuracy: 0.8677 - val_loss: 0.5084 - val_accuracy: 0.8341
Epoch 90/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4115 - accuracy: 0.8601 - val_loss: 0.4140 - val_accuracy: 0.8575
Epoch 91/200
128/128 [==============================] - 5s 43ms/step - loss: 0.4168 - accuracy: 0.8609 - val_loss: 0.9081 - val_accuracy: 0.7453
Epoch 92/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4270 - accuracy: 0.8562 - val_loss: 0.4540 - val_accuracy: 0.8621
Epoch 93/200
128/128 [==============================] - 5s 42ms/step - loss: 0.4080 - accuracy: 0.8567 - val_loss: 1.3221 - val_accuracy: 0.6589
Epoch 94/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4159 - accuracy: 0.8556 - val_loss: 0.4579 - val_accuracy: 0.8341
Epoch 95/200
128/128 [==============================] - 5s 43ms/step - loss: 0.3769 - accuracy: 0.8750 - val_loss: 0.4313 - val_accuracy: 0.8692
Epoch 96/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4278 - accuracy: 0.8538 - val_loss: 1.0849 - val_accuracy: 0.7196
Epoch 97/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4084 - accuracy: 0.8522 - val_loss: 0.6176 - val_accuracy: 0.8131
Epoch 98/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3882 - accuracy: 0.8708 - val_loss: 0.4960 - val_accuracy: 0.8364
Epoch 99/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4020 - accuracy: 0.8638 - val_loss: 0.7156 - val_accuracy: 0.8061
Epoch 100/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4031 - accuracy: 0.8674 - val_loss: 1.2692 - val_accuracy: 0.6308
Epoch 101/200
128/128 [==============================] - 5s 41ms/step - loss: 0.4076 - accuracy: 0.8617 - val_loss: 0.6873 - val_accuracy: 0.8061
Epoch 102/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3887 - accuracy: 0.8577 - val_loss: 1.3242 - val_accuracy: 0.7103
Epoch 103/200
128/128 [==============================] - 6s 43ms/step - loss: 0.4041 - accuracy: 0.8583 - val_loss: 0.3587 - val_accuracy: 0.8715
Epoch 104/200
128/128 [==============================] - 5s 42ms/step - loss: 0.3864 - accuracy: 0.8666 - val_loss: 4.7652 - val_accuracy: 0.3692
Epoch 105/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3895 - accuracy: 0.8680 - val_loss: 0.4881 - val_accuracy: 0.8528
Epoch 106/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3559 - accuracy: 0.8805 - val_loss: 0.4824 - val_accuracy: 0.8435
Epoch 107/200
128/128 [==============================] - 6s 43ms/step - loss: 0.3943 - accuracy: 0.8648 - val_loss: 1.5649 - val_accuracy: 0.5958
Epoch 108/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3966 - accuracy: 0.8617 - val_loss: 1.5344 - val_accuracy: 0.6939
Epoch 109/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3637 - accuracy: 0.8742 - val_loss: 1.0776 - val_accuracy: 0.7547
Epoch 110/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3479 - accuracy: 0.8792 - val_loss: 0.6281 - val_accuracy: 0.8037
Epoch 111/200
128/128 [==============================] - 6s 44ms/step - loss: 0.3498 - accuracy: 0.8863 - val_loss: 1.5955 - val_accuracy: 0.6659
Epoch 112/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3706 - accuracy: 0.8719 - val_loss: 0.3286 - val_accuracy: 0.9206
Epoch 113/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3658 - accuracy: 0.8750 - val_loss: 0.4361 - val_accuracy: 0.8692
Epoch 114/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3485 - accuracy: 0.8818 - val_loss: 0.6118 - val_accuracy: 0.8505
Epoch 115/200
128/128 [==============================] - 5s 42ms/step - loss: 0.3708 - accuracy: 0.8842 - val_loss: 0.3111 - val_accuracy: 0.8879
Epoch 116/200
128/128 [==============================] - 5s 43ms/step - loss: 0.3295 - accuracy: 0.8910 - val_loss: 1.2392 - val_accuracy: 0.6285
Epoch 117/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3717 - accuracy: 0.8750 - val_loss: 0.5155 - val_accuracy: 0.8621
Epoch 118/200
128/128 [==============================] - 6s 43ms/step - loss: 0.3630 - accuracy: 0.8805 - val_loss: 2.9767 - val_accuracy: 0.3715
Epoch 119/200
128/128 [==============================] - 7s 51ms/step - loss: 0.3402 - accuracy: 0.8784 - val_loss: 2.2329 - val_accuracy: 0.4836
Epoch 120/200
128/128 [==============================] - 5s 42ms/step - loss: 0.3803 - accuracy: 0.8763 - val_loss: 1.1495 - val_accuracy: 0.7220
Epoch 121/200
128/128 [==============================] - 5s 42ms/step - loss: 0.3283 - accuracy: 0.8908 - val_loss: 0.2934 - val_accuracy: 0.9065
Epoch 122/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3330 - accuracy: 0.8887 - val_loss: 2.3880 - val_accuracy: 0.5234
Epoch 123/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3398 - accuracy: 0.8818 - val_loss: 1.3506 - val_accuracy: 0.7196
Epoch 124/200
128/128 [==============================] - 5s 42ms/step - loss: 0.3437 - accuracy: 0.8839 - val_loss: 0.5650 - val_accuracy: 0.8411
Epoch 125/200
128/128 [==============================] - 5s 42ms/step - loss: 0.3481 - accuracy: 0.8795 - val_loss: 0.8538 - val_accuracy: 0.7570
Epoch 126/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3364 - accuracy: 0.8884 - val_loss: 1.0054 - val_accuracy: 0.7547
Epoch 127/200
128/128 [==============================] - 6s 43ms/step - loss: 0.3277 - accuracy: 0.8863 - val_loss: 0.8878 - val_accuracy: 0.7757
Epoch 128/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3419 - accuracy: 0.8871 - val_loss: 0.8952 - val_accuracy: 0.7850
Epoch 129/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3381 - accuracy: 0.8879 - val_loss: 1.2390 - val_accuracy: 0.7243
Epoch 130/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3284 - accuracy: 0.8923 - val_loss: 0.4740 - val_accuracy: 0.8692
Epoch 131/200
128/128 [==============================] - 6s 44ms/step - loss: 0.3294 - accuracy: 0.8908 - val_loss: 0.4837 - val_accuracy: 0.8832
Epoch 132/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3214 - accuracy: 0.8957 - val_loss: 0.4135 - val_accuracy: 0.8785
Epoch 133/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3125 - accuracy: 0.8931 - val_loss: 0.3945 - val_accuracy: 0.8762
Epoch 134/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3208 - accuracy: 0.8915 - val_loss: 2.0014 - val_accuracy: 0.4579
Epoch 135/200
128/128 [==============================] - 5s 42ms/step - loss: 0.3952 - accuracy: 0.8687 - val_loss: 0.4472 - val_accuracy: 0.8458
Epoch 136/200
128/128 [==============================] - 5s 42ms/step - loss: 0.3160 - accuracy: 0.8902 - val_loss: 0.3148 - val_accuracy: 0.9136
Epoch 137/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3193 - accuracy: 0.8936 - val_loss: 0.4318 - val_accuracy: 0.8855
Epoch 138/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3251 - accuracy: 0.8894 - val_loss: 0.3365 - val_accuracy: 0.8949
Epoch 139/200
128/128 [==============================] - 6s 43ms/step - loss: 0.3138 - accuracy: 0.8970 - val_loss: 0.8588 - val_accuracy: 0.7570
Epoch 140/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2957 - accuracy: 0.9002 - val_loss: 0.7345 - val_accuracy: 0.8061
Epoch 141/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2802 - accuracy: 0.9062 - val_loss: 0.8527 - val_accuracy: 0.7266
Epoch 142/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3011 - accuracy: 0.8965 - val_loss: 0.9023 - val_accuracy: 0.7173
Epoch 143/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2907 - accuracy: 0.9046 - val_loss: 0.6603 - val_accuracy: 0.8388
Epoch 144/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2945 - accuracy: 0.8947 - val_loss: 0.3574 - val_accuracy: 0.8972
Epoch 145/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3221 - accuracy: 0.8887 - val_loss: 0.3667 - val_accuracy: 0.8785
Epoch 146/200
128/128 [==============================] - 5s 43ms/step - loss: 0.3186 - accuracy: 0.8926 - val_loss: 0.4729 - val_accuracy: 0.8458
Epoch 147/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3063 - accuracy: 0.8973 - val_loss: 0.8123 - val_accuracy: 0.8318
Epoch 148/200
128/128 [==============================] - 6s 47ms/step - loss: 0.2944 - accuracy: 0.9004 - val_loss: 0.7429 - val_accuracy: 0.7897
Epoch 149/200
128/128 [==============================] - 6s 46ms/step - loss: 0.3097 - accuracy: 0.8913 - val_loss: 0.3475 - val_accuracy: 0.8879
Epoch 150/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2879 - accuracy: 0.8968 - val_loss: 0.3697 - val_accuracy: 0.8762
Epoch 151/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2880 - accuracy: 0.8986 - val_loss: 1.0873 - val_accuracy: 0.7126
Epoch 152/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2981 - accuracy: 0.8999 - val_loss: 0.4143 - val_accuracy: 0.8738
Epoch 153/200
128/128 [==============================] - 6s 43ms/step - loss: 0.2985 - accuracy: 0.8986 - val_loss: 0.5909 - val_accuracy: 0.8435
Epoch 154/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2989 - accuracy: 0.9023 - val_loss: 0.6861 - val_accuracy: 0.8598
Epoch 155/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2899 - accuracy: 0.8978 - val_loss: 0.4657 - val_accuracy: 0.8832
Epoch 156/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2952 - accuracy: 0.9041 - val_loss: 0.6364 - val_accuracy: 0.7991
Epoch 157/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2725 - accuracy: 0.9041 - val_loss: 0.6693 - val_accuracy: 0.8318
Epoch 158/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3070 - accuracy: 0.9044 - val_loss: 0.5411 - val_accuracy: 0.8341
Epoch 159/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2750 - accuracy: 0.9088 - val_loss: 0.4746 - val_accuracy: 0.8645
Epoch 160/200
128/128 [==============================] - 6s 43ms/step - loss: 0.3074 - accuracy: 0.8965 - val_loss: 0.5492 - val_accuracy: 0.8364
Epoch 161/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2669 - accuracy: 0.9141 - val_loss: 1.0586 - val_accuracy: 0.7290
Epoch 162/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3084 - accuracy: 0.8955 - val_loss: 0.4879 - val_accuracy: 0.8551
Epoch 163/200
128/128 [==============================] - 5s 41ms/step - loss: 0.3011 - accuracy: 0.8970 - val_loss: 0.7613 - val_accuracy: 0.7921
Epoch 164/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2828 - accuracy: 0.9088 - val_loss: 2.5852 - val_accuracy: 0.5304
Epoch 165/200
128/128 [==============================] - 5s 43ms/step - loss: 0.2863 - accuracy: 0.9088 - val_loss: 0.4113 - val_accuracy: 0.8785
Epoch 166/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2936 - accuracy: 0.9075 - val_loss: 0.5468 - val_accuracy: 0.8318
Epoch 167/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2875 - accuracy: 0.9054 - val_loss: 0.4999 - val_accuracy: 0.8575
Epoch 168/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2408 - accuracy: 0.9180 - val_loss: 0.3272 - val_accuracy: 0.9019
Epoch 169/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2539 - accuracy: 0.9198 - val_loss: 0.4010 - val_accuracy: 0.8738
Epoch 170/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2727 - accuracy: 0.9101 - val_loss: 0.7971 - val_accuracy: 0.8224
Epoch 171/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2811 - accuracy: 0.9012 - val_loss: 3.8131 - val_accuracy: 0.4743
Epoch 172/200
128/128 [==============================] - 6s 44ms/step - loss: 0.2796 - accuracy: 0.9054 - val_loss: 0.7114 - val_accuracy: 0.8411
Epoch 173/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2725 - accuracy: 0.9088 - val_loss: 0.5974 - val_accuracy: 0.8598
Epoch 174/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2513 - accuracy: 0.9167 - val_loss: 0.3013 - val_accuracy: 0.9019
Epoch 175/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2717 - accuracy: 0.9073 - val_loss: 0.6183 - val_accuracy: 0.8154
Epoch 176/200
128/128 [==============================] - 6s 43ms/step - loss: 0.2841 - accuracy: 0.9073 - val_loss: 0.5470 - val_accuracy: 0.8621
Epoch 177/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2545 - accuracy: 0.9149 - val_loss: 0.3834 - val_accuracy: 0.8832
Epoch 178/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2860 - accuracy: 0.9010 - val_loss: 1.3895 - val_accuracy: 0.7383
Epoch 179/200
128/128 [==============================] - 7s 53ms/step - loss: 0.2698 - accuracy: 0.9101 - val_loss: 0.4915 - val_accuracy: 0.8692
Epoch 180/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2815 - accuracy: 0.9036 - val_loss: 0.6526 - val_accuracy: 0.8294
Epoch 181/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2644 - accuracy: 0.9149 - val_loss: 0.4148 - val_accuracy: 0.8832
Epoch 182/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2735 - accuracy: 0.9086 - val_loss: 0.3858 - val_accuracy: 0.8925
Epoch 183/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2686 - accuracy: 0.9112 - val_loss: 0.8041 - val_accuracy: 0.7780
Epoch 184/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2589 - accuracy: 0.9141 - val_loss: 0.4944 - val_accuracy: 0.8715
Epoch 185/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2599 - accuracy: 0.9172 - val_loss: 0.4712 - val_accuracy: 0.8715
Epoch 186/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2653 - accuracy: 0.9094 - val_loss: 0.3401 - val_accuracy: 0.8995
Epoch 187/200
128/128 [==============================] - 6s 44ms/step - loss: 0.2734 - accuracy: 0.9099 - val_loss: 0.3458 - val_accuracy: 0.9112
Epoch 188/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2477 - accuracy: 0.9206 - val_loss: 0.3527 - val_accuracy: 0.9136
Epoch 189/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2655 - accuracy: 0.9133 - val_loss: 0.4563 - val_accuracy: 0.8855
Epoch 190/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2699 - accuracy: 0.9130 - val_loss: 0.4260 - val_accuracy: 0.8832
Epoch 191/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2661 - accuracy: 0.9109 - val_loss: 0.5272 - val_accuracy: 0.8762
Epoch 192/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2636 - accuracy: 0.9080 - val_loss: 0.7638 - val_accuracy: 0.7804
Epoch 193/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2402 - accuracy: 0.9206 - val_loss: 0.2480 - val_accuracy: 0.9299
Epoch 194/200
128/128 [==============================] - 6s 43ms/step - loss: 0.2631 - accuracy: 0.9112 - val_loss: 1.0812 - val_accuracy: 0.7079
Epoch 195/200
128/128 [==============================] - 5s 43ms/step - loss: 0.2570 - accuracy: 0.9143 - val_loss: 0.4845 - val_accuracy: 0.8481
Epoch 196/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2503 - accuracy: 0.9217 - val_loss: 0.2763 - val_accuracy: 0.9182
Epoch 197/200
128/128 [==============================] - 5s 42ms/step - loss: 0.2310 - accuracy: 0.9209 - val_loss: 0.4874 - val_accuracy: 0.8808
Epoch 198/200
128/128 [==============================] - 6s 43ms/step - loss: 0.2703 - accuracy: 0.9167 - val_loss: 0.4553 - val_accuracy: 0.8902
Epoch 199/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2219 - accuracy: 0.9274 - val_loss: 0.4491 - val_accuracy: 0.8668
Epoch 200/200
128/128 [==============================] - 5s 41ms/step - loss: 0.2278 - accuracy: 0.9230 - val_loss: 0.3396 - val_accuracy: 0.9229

Model Evaluation

plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()

Observations:

• We can see from the above plot that the training accuracy of the model and the validation accuracy are very good.
• The Training and Validation accuracy scores are well aligned
• The model seems to be having a good fit without overfitting
• This seems to be a good model to predict the images

Model Evaluation on Training dataset

model.evaluate(X_train_normalized, y_train_encoded, verbose=1)
model2_TrainDF = model_performance_classification_sklearn_with_threshold(
    model, X_train_normalized, y_train_encoded,
)
model2_TrainDF

121/121 [==============================] - 1s 7ms/step - loss: 0.1691 - accuracy: 0.9465
121/121 [==============================] - 1s 6ms/step

	Accuracy	Recall	Precision	F1
0	0.946452	0.946452	0.947343	0.946507

Observations

• The Training accuracy of this model is 94.6% and the other metrics also are in the same range. It has a good accuracy score
• We can observe that most of the classes are predicted correctly
• Except for the classes 0, 3, 6 & 9, all other classes are well classified.

Model Evaluation on Validation dataset

model.evaluate(X_val_normalized, y_val_encoded, verbose=1)
model2_ValDF = model_performance_classification_sklearn_with_threshold(
    model, X_val_normalized,y_val_encoded,
)
model2_ValDF

14/14 [==============================] - 0s 7ms/step - loss: 0.3396 - accuracy: 0.9229
14/14 [==============================] - 0s 6ms/step

	Accuracy	Recall	Precision	F1
0	0.922897	0.922897	0.927033	0.923229

Observations

• The Validation accuracy of this model is 92.3% and the other metrics also are in the same range. Precision has a higher metric
• When compared with the Training accurace, there is only ~4% variance in the Accuracy scores. The training and validation data seems to be fitting very well in this model

Model Evaluation on Test dataset

model.evaluate(X_test_normalized, y_test_encoded, verbose=2)
model2_TestDF = model_performance_classification_sklearn_with_threshold(model, X_test_normalized, y_test_encoded)
model2_TestDF

15/15 - 0s - loss: 0.3019 - accuracy: 0.9116 - 97ms/epoch - 6ms/step
15/15 [==============================] - 0s 6ms/step

	Accuracy	Recall	Precision	F1
0	0.911579	0.911579	0.914286	0.91229

Observations

• We can observe that the accuracy has improved compared to our previous models and outperformed all other models so far.
• The Validation accuracy of this model is 91.1% and the other metrics also are in the same range. The accuracy is very good and also the F1 score has a higher metric which is very good. This would help to determine both positive & negative predictions more correctly
• When compared with the Training accuracy, there is only ~3% variance in the Accuracy scores. The training and testing data seems to be fitting very well in this model

Visualizing the Predictions:

# Visualizing the predicted and correct label of images from test data
rows = 1                                                                        
cols = 5                                                                        
for i in range(cols):
    for j in range(rows):
      random_index = np.random.randint(0, 400)    
      plt.imshow(X_test[random_index])
      plt.show()

      print('Predicted Label', enc.inverse_transform(model.predict((X_test_normalized[random_index].reshape(1,64,64,3)))))   
      print('True Label', enc.inverse_transform(y_test_encoded)[random_index])     

1/1 [==============================] - 0s 20ms/step
Predicted Label ['Common wheat']
True Label Common wheat

1/1 [==============================] - 0s 18ms/step
Predicted Label ['Maize']
True Label Maize

1/1 [==============================] - 0s 17ms/step
Predicted Label ['Scentless Mayweed']
True Label Scentless Mayweed

1/1 [==============================] - 0s 18ms/step
Predicted Label ['Small-flowered Cranesbill']
True Label Small-flowered Cranesbill

1/1 [==============================] - 0s 18ms/step
Predicted Label ['Shepherds Purse']
True Label Shepherds Purse

Inference:

• The random images selected has the correct prediction so far in this model.

Model # 3 - CNN with Transfer Learning using VGG16

---

• Creating a CNN model by leveraging an existing model VGG16 for Transfer learning

# Clearing backend
backend.clear_session()

# Fixing the seed for random number generators
import random
np.random.seed(42)
random.seed(42)
tf.random.set_seed(42)

Using VGG16 model for the Transfer Learing

# Applying the VGG16 model
from tensorflow.keras.models import Model
from keras.applications.vgg16 import VGG16

vgg_model = VGG16(weights='imagenet', include_top = False, input_shape = (64,64,3))
vgg_model.summary()

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/vgg16/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5
58889256/58889256 [==============================] - 2s 0us/step
Model: "vgg16"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_1 (InputLayer)        [(None, 64, 64, 3)]       0         
                                                                 
 block1_conv1 (Conv2D)       (None, 64, 64, 64)        1792      
                                                                 
 block1_conv2 (Conv2D)       (None, 64, 64, 64)        36928     
                                                                 
 block1_pool (MaxPooling2D)  (None, 32, 32, 64)        0         
                                                                 
 block2_conv1 (Conv2D)       (None, 32, 32, 128)       73856     
                                                                 
 block2_conv2 (Conv2D)       (None, 32, 32, 128)       147584    
                                                                 
 block2_pool (MaxPooling2D)  (None, 16, 16, 128)       0         
                                                                 
 block3_conv1 (Conv2D)       (None, 16, 16, 256)       295168    
                                                                 
 block3_conv2 (Conv2D)       (None, 16, 16, 256)       590080    
                                                                 
 block3_conv3 (Conv2D)       (None, 16, 16, 256)       590080    
                                                                 
 block3_pool (MaxPooling2D)  (None, 8, 8, 256)         0         
                                                                 
 block4_conv1 (Conv2D)       (None, 8, 8, 512)         1180160   
                                                                 
 block4_conv2 (Conv2D)       (None, 8, 8, 512)         2359808   
                                                                 
 block4_conv3 (Conv2D)       (None, 8, 8, 512)         2359808   
                                                                 
 block4_pool (MaxPooling2D)  (None, 4, 4, 512)         0         
                                                                 
 block5_conv1 (Conv2D)       (None, 4, 4, 512)         2359808   
                                                                 
 block5_conv2 (Conv2D)       (None, 4, 4, 512)         2359808   
                                                                 
 block5_conv3 (Conv2D)       (None, 4, 4, 512)         2359808   
                                                                 
 block5_pool (MaxPooling2D)  (None, 2, 2, 512)         0         
                                                                 
=================================================================
Total params: 14,714,688
Trainable params: 14,714,688
Non-trainable params: 0
_________________________________________________________________

Observations:

• There are a total of 14,714,688 params with 14,714,688 trainable params

# Making all the layers of the VGG model non-trainable. i.e. freezing them
for layer in vgg_model.layers:
    layer.trainable = False

new_model = Sequential()

# Adding the convolutional part of the VGG16 model from above
new_model.add(vgg_model)

# Flattening the output of the VGG16 model because it is from a convolutional layer
new_model.add(Flatten())

# Adding hidden layer with 64 neurons, relu as activation function and, he_normal as weight initializer.
new_model.add(Dense(64, activation="relu", kernel_initializer="he_normal", ))
new_model.add(Dense(64, activation="relu", kernel_initializer="he_normal", ))
new_model.add(BatchNormalization())

# Adding hidden layer with 32 neurons, relu as activation function and, he_normal as weight initializer
new_model.add(Dense(32, activation="relu", kernel_initializer="he_normal",))

# Adding Dropout with 20%
new_model.add(Dropout(0.2))

# Adding hidden layer with 16 neurons, relu as activation function and, he_normal as weight initializer
new_model.add(Dense(32, activation="relu", kernel_initializer="he_normal",))
new_model.add(Dense(16, activation="relu", kernel_initializer="he_normal",))
new_model.add(BatchNormalization())

# Adding Dropout with 20%
new_model.add(Dropout(0.2))

# Adding hidden layer with 8 neurons, relu as activation function and, he_normal as weight initializer
new_model.add(Dense(8, activation="relu", kernel_initializer="he_normal",))
new_model.add(BatchNormalization())

new_model.add(Dense(16, activation='relu'))
new_model.add(Dense(12, activation='softmax'))

opt=Adam()
# Compile model
new_model.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])

# Generating the summary of the model
new_model.summary()

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 vgg16 (Functional)          (None, 2, 2, 512)         14714688  
                                                                 
 flatten (Flatten)           (None, 2048)              0         
                                                                 
 dense (Dense)               (None, 64)                131136    
                                                                 
 dense_1 (Dense)             (None, 64)                4160      
                                                                 
 batch_normalization (BatchN  (None, 64)               256       
 ormalization)                                                   
                                                                 
 dense_2 (Dense)             (None, 32)                2080      
                                                                 
 dropout (Dropout)           (None, 32)                0         
                                                                 
 dense_3 (Dense)             (None, 32)                1056      
                                                                 
 dense_4 (Dense)             (None, 16)                528       
                                                                 
 batch_normalization_1 (Batc  (None, 16)               64        
 hNormalization)                                                 
                                                                 
 dropout_1 (Dropout)         (None, 16)                0         
                                                                 
 dense_5 (Dense)             (None, 8)                 136       
                                                                 
 batch_normalization_2 (Batc  (None, 8)                32        
 hNormalization)                                                 
                                                                 
 dense_6 (Dense)             (None, 16)                144       
                                                                 
 dense_7 (Dense)             (None, 12)                204       
                                                                 
=================================================================
Total params: 14,854,484
Trainable params: 139,620
Non-trainable params: 14,714,864
_________________________________________________________________

Observations:

• There are a total of 14,854,484 params with 139,620 trainable params. The remaining params are either frozen or not applied
• We have used VGG16 for intitial learning and applied the layers on top of it
• We are using the loss function - categorical_crossentropy as this is a multi-class classification problem and Adam as the optimizer
• Also, we are choosing accuracy as the metric to measure the performance of the model

Fitting the model on the train data

• Applying the VGG model along with Data Augumentation, Epochs & Batch Size parameters.

# Epochs
epochs = 200
# Batch size
batch_size = 30


history_vgg16 = new_model.fit(train_datagen.flow(X_train_normalized,y_train_encoded,
                                       batch_size=batch_size, 
                                       seed=42,
                                       shuffle=False),
                    epochs=epochs,
                    steps_per_epoch=X_train_normalized.shape[0] // batch_size,
                    validation_data=(X_val_normalized,y_val_encoded),
                    verbose=1)

Epoch 1/200
128/128 [==============================] - 9s 54ms/step - loss: 2.4564 - accuracy: 0.1462 - val_loss: 2.4526 - val_accuracy: 0.1472
Epoch 2/200
128/128 [==============================] - 6s 45ms/step - loss: 2.2239 - accuracy: 0.2478 - val_loss: 2.1505 - val_accuracy: 0.2477
Epoch 3/200
128/128 [==============================] - 6s 43ms/step - loss: 2.0379 - accuracy: 0.2981 - val_loss: 2.0983 - val_accuracy: 0.2874
Epoch 4/200
128/128 [==============================] - 6s 44ms/step - loss: 1.8941 - accuracy: 0.3128 - val_loss: 1.8832 - val_accuracy: 0.3224
Epoch 5/200
128/128 [==============================] - 6s 45ms/step - loss: 1.8254 - accuracy: 0.3353 - val_loss: 1.7882 - val_accuracy: 0.3458
Epoch 6/200
128/128 [==============================] - 6s 43ms/step - loss: 1.7612 - accuracy: 0.3642 - val_loss: 1.8407 - val_accuracy: 0.3738
Epoch 7/200
128/128 [==============================] - 6s 48ms/step - loss: 1.7381 - accuracy: 0.3592 - val_loss: 1.6876 - val_accuracy: 0.4112
Epoch 8/200
128/128 [==============================] - 6s 43ms/step - loss: 1.7040 - accuracy: 0.3815 - val_loss: 1.7948 - val_accuracy: 0.3762
Epoch 9/200
128/128 [==============================] - 6s 44ms/step - loss: 1.6445 - accuracy: 0.4040 - val_loss: 1.6377 - val_accuracy: 0.4416
Epoch 10/200
128/128 [==============================] - 6s 43ms/step - loss: 1.6354 - accuracy: 0.4139 - val_loss: 1.7611 - val_accuracy: 0.3902
Epoch 11/200
128/128 [==============================] - 6s 43ms/step - loss: 1.6256 - accuracy: 0.4142 - val_loss: 1.5674 - val_accuracy: 0.4463
Epoch 12/200
128/128 [==============================] - 6s 43ms/step - loss: 1.6069 - accuracy: 0.4218 - val_loss: 1.9095 - val_accuracy: 0.3528
Epoch 13/200
128/128 [==============================] - 6s 48ms/step - loss: 1.5727 - accuracy: 0.4338 - val_loss: 2.1961 - val_accuracy: 0.2944
Epoch 14/200
128/128 [==============================] - 6s 43ms/step - loss: 1.5582 - accuracy: 0.4383 - val_loss: 1.5683 - val_accuracy: 0.4393
Epoch 15/200
128/128 [==============================] - 6s 43ms/step - loss: 1.5264 - accuracy: 0.4532 - val_loss: 1.5091 - val_accuracy: 0.4626
Epoch 16/200
128/128 [==============================] - 6s 43ms/step - loss: 1.5253 - accuracy: 0.4566 - val_loss: 1.5643 - val_accuracy: 0.4486
Epoch 17/200
128/128 [==============================] - 6s 45ms/step - loss: 1.5126 - accuracy: 0.4697 - val_loss: 1.7227 - val_accuracy: 0.4112
Epoch 18/200
128/128 [==============================] - 5s 43ms/step - loss: 1.5238 - accuracy: 0.4697 - val_loss: 1.5137 - val_accuracy: 0.4603
Epoch 19/200
128/128 [==============================] - 6s 43ms/step - loss: 1.5148 - accuracy: 0.4650 - val_loss: 1.7024 - val_accuracy: 0.3598
Epoch 20/200
128/128 [==============================] - 6s 44ms/step - loss: 1.5033 - accuracy: 0.4676 - val_loss: 1.9007 - val_accuracy: 0.3294
Epoch 21/200
128/128 [==============================] - 5s 43ms/step - loss: 1.4784 - accuracy: 0.4737 - val_loss: 1.5038 - val_accuracy: 0.4696
Epoch 22/200
128/128 [==============================] - 6s 44ms/step - loss: 1.4577 - accuracy: 0.4881 - val_loss: 1.5272 - val_accuracy: 0.4416
Epoch 23/200
128/128 [==============================] - 6s 44ms/step - loss: 1.4410 - accuracy: 0.4910 - val_loss: 1.5277 - val_accuracy: 0.4813
Epoch 24/200
128/128 [==============================] - 6s 44ms/step - loss: 1.4568 - accuracy: 0.4813 - val_loss: 1.5773 - val_accuracy: 0.4556
Epoch 25/200
128/128 [==============================] - 6s 45ms/step - loss: 1.4391 - accuracy: 0.4928 - val_loss: 1.4457 - val_accuracy: 0.5210
Epoch 26/200
128/128 [==============================] - 6s 44ms/step - loss: 1.4379 - accuracy: 0.4952 - val_loss: 1.3431 - val_accuracy: 0.5514
Epoch 27/200
128/128 [==============================] - 6s 44ms/step - loss: 1.4281 - accuracy: 0.4975 - val_loss: 2.2969 - val_accuracy: 0.3248
Epoch 28/200
128/128 [==============================] - 6s 46ms/step - loss: 1.4068 - accuracy: 0.5014 - val_loss: 1.3634 - val_accuracy: 0.5257
Epoch 29/200
128/128 [==============================] - 6s 43ms/step - loss: 1.4069 - accuracy: 0.5064 - val_loss: 1.9899 - val_accuracy: 0.3668
Epoch 30/200
128/128 [==============================] - 6s 45ms/step - loss: 1.3893 - accuracy: 0.5088 - val_loss: 1.3957 - val_accuracy: 0.5117
Epoch 31/200
128/128 [==============================] - 6s 43ms/step - loss: 1.3830 - accuracy: 0.5117 - val_loss: 1.2976 - val_accuracy: 0.5514
Epoch 32/200
128/128 [==============================] - 6s 46ms/step - loss: 1.3750 - accuracy: 0.5187 - val_loss: 1.4194 - val_accuracy: 0.5210
Epoch 33/200
128/128 [==============================] - 6s 43ms/step - loss: 1.3633 - accuracy: 0.5216 - val_loss: 1.4184 - val_accuracy: 0.4953
Epoch 34/200
128/128 [==============================] - 6s 45ms/step - loss: 1.3709 - accuracy: 0.5140 - val_loss: 1.7312 - val_accuracy: 0.4369
Epoch 35/200
128/128 [==============================] - 6s 44ms/step - loss: 1.3577 - accuracy: 0.5185 - val_loss: 1.4089 - val_accuracy: 0.5023
Epoch 36/200
128/128 [==============================] - 6s 45ms/step - loss: 1.3321 - accuracy: 0.5360 - val_loss: 1.4083 - val_accuracy: 0.5023
Epoch 37/200
128/128 [==============================] - 6s 44ms/step - loss: 1.3468 - accuracy: 0.5269 - val_loss: 1.4917 - val_accuracy: 0.4766
Epoch 38/200
128/128 [==============================] - 7s 54ms/step - loss: 1.3270 - accuracy: 0.5413 - val_loss: 1.3918 - val_accuracy: 0.5070
Epoch 39/200
128/128 [==============================] - 6s 44ms/step - loss: 1.3362 - accuracy: 0.5331 - val_loss: 1.4782 - val_accuracy: 0.4673
Epoch 40/200
128/128 [==============================] - 6s 44ms/step - loss: 1.3084 - accuracy: 0.5436 - val_loss: 1.3914 - val_accuracy: 0.5210
Epoch 41/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2927 - accuracy: 0.5468 - val_loss: 1.3751 - val_accuracy: 0.5234
Epoch 42/200
128/128 [==============================] - 6s 45ms/step - loss: 1.3022 - accuracy: 0.5368 - val_loss: 1.4383 - val_accuracy: 0.4766
Epoch 43/200
128/128 [==============================] - 6s 44ms/step - loss: 1.3077 - accuracy: 0.5358 - val_loss: 1.2781 - val_accuracy: 0.5561
Epoch 44/200
128/128 [==============================] - 6s 43ms/step - loss: 1.3005 - accuracy: 0.5447 - val_loss: 1.3189 - val_accuracy: 0.5280
Epoch 45/200
128/128 [==============================] - 6s 43ms/step - loss: 1.2761 - accuracy: 0.5520 - val_loss: 1.2953 - val_accuracy: 0.5467
Epoch 46/200
128/128 [==============================] - 6s 43ms/step - loss: 1.2893 - accuracy: 0.5572 - val_loss: 1.3069 - val_accuracy: 0.5537
Epoch 47/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2744 - accuracy: 0.5499 - val_loss: 1.3804 - val_accuracy: 0.4860
Epoch 48/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2589 - accuracy: 0.5538 - val_loss: 1.5662 - val_accuracy: 0.4579
Epoch 49/200
128/128 [==============================] - 6s 43ms/step - loss: 1.2826 - accuracy: 0.5614 - val_loss: 1.5342 - val_accuracy: 0.4790
Epoch 50/200
128/128 [==============================] - 6s 43ms/step - loss: 1.2610 - accuracy: 0.5586 - val_loss: 1.3293 - val_accuracy: 0.5187
Epoch 51/200
128/128 [==============================] - 6s 45ms/step - loss: 1.2759 - accuracy: 0.5523 - val_loss: 1.3642 - val_accuracy: 0.5093
Epoch 52/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2666 - accuracy: 0.5523 - val_loss: 1.2694 - val_accuracy: 0.5561
Epoch 53/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2571 - accuracy: 0.5672 - val_loss: 1.5299 - val_accuracy: 0.4977
Epoch 54/200
128/128 [==============================] - 6s 43ms/step - loss: 1.2428 - accuracy: 0.5680 - val_loss: 1.2175 - val_accuracy: 0.5607
Epoch 55/200
128/128 [==============================] - 6s 45ms/step - loss: 1.2528 - accuracy: 0.5546 - val_loss: 1.1794 - val_accuracy: 0.5864
Epoch 56/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2463 - accuracy: 0.5648 - val_loss: 1.4641 - val_accuracy: 0.4813
Epoch 57/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2606 - accuracy: 0.5669 - val_loss: 1.3700 - val_accuracy: 0.5444
Epoch 58/200
128/128 [==============================] - 6s 45ms/step - loss: 1.2252 - accuracy: 0.5711 - val_loss: 1.2054 - val_accuracy: 0.5771
Epoch 59/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2211 - accuracy: 0.5753 - val_loss: 1.3824 - val_accuracy: 0.5210
Epoch 60/200
128/128 [==============================] - 6s 47ms/step - loss: 1.2442 - accuracy: 0.5769 - val_loss: 1.8047 - val_accuracy: 0.4276
Epoch 61/200
128/128 [==============================] - 6s 46ms/step - loss: 1.2462 - accuracy: 0.5701 - val_loss: 1.3206 - val_accuracy: 0.5584
Epoch 62/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2416 - accuracy: 0.5643 - val_loss: 1.2202 - val_accuracy: 0.5864
Epoch 63/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2232 - accuracy: 0.5790 - val_loss: 1.7144 - val_accuracy: 0.4579
Epoch 64/200
128/128 [==============================] - 6s 45ms/step - loss: 1.2171 - accuracy: 0.5751 - val_loss: 1.2867 - val_accuracy: 0.5584
Epoch 65/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2163 - accuracy: 0.5795 - val_loss: 1.1896 - val_accuracy: 0.5981
Epoch 66/200
128/128 [==============================] - 6s 45ms/step - loss: 1.2062 - accuracy: 0.5790 - val_loss: 1.2001 - val_accuracy: 0.5631
Epoch 67/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2262 - accuracy: 0.5779 - val_loss: 1.3772 - val_accuracy: 0.5187
Epoch 68/200
128/128 [==============================] - 6s 46ms/step - loss: 1.2188 - accuracy: 0.5808 - val_loss: 1.2108 - val_accuracy: 0.5958
Epoch 69/200
128/128 [==============================] - 7s 55ms/step - loss: 1.1897 - accuracy: 0.5813 - val_loss: 1.1840 - val_accuracy: 0.5958
Epoch 70/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1874 - accuracy: 0.5937 - val_loss: 1.4988 - val_accuracy: 0.4836
Epoch 71/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1899 - accuracy: 0.5876 - val_loss: 1.1601 - val_accuracy: 0.6098
Epoch 72/200
128/128 [==============================] - 6s 45ms/step - loss: 1.2144 - accuracy: 0.5808 - val_loss: 1.2208 - val_accuracy: 0.5678
Epoch 73/200
128/128 [==============================] - 6s 44ms/step - loss: 1.2016 - accuracy: 0.5795 - val_loss: 1.2039 - val_accuracy: 0.5841
Epoch 74/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1791 - accuracy: 0.5965 - val_loss: 1.7227 - val_accuracy: 0.4229
Epoch 75/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1814 - accuracy: 0.5910 - val_loss: 1.2938 - val_accuracy: 0.5561
Epoch 76/200
128/128 [==============================] - 6s 45ms/step - loss: 1.1705 - accuracy: 0.5992 - val_loss: 1.2303 - val_accuracy: 0.5724
Epoch 77/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1988 - accuracy: 0.5808 - val_loss: 1.1627 - val_accuracy: 0.6168
Epoch 78/200
128/128 [==============================] - 6s 45ms/step - loss: 1.1776 - accuracy: 0.5903 - val_loss: 1.2383 - val_accuracy: 0.5678
Epoch 79/200
128/128 [==============================] - 6s 43ms/step - loss: 1.1895 - accuracy: 0.5874 - val_loss: 1.1462 - val_accuracy: 0.5864
Epoch 80/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1776 - accuracy: 0.5939 - val_loss: 1.2173 - val_accuracy: 0.5888
Epoch 81/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1741 - accuracy: 0.6049 - val_loss: 1.2303 - val_accuracy: 0.5748
Epoch 82/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1853 - accuracy: 0.5866 - val_loss: 1.2542 - val_accuracy: 0.5514
Epoch 83/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1685 - accuracy: 0.5973 - val_loss: 1.3804 - val_accuracy: 0.5491
Epoch 84/200
128/128 [==============================] - 6s 45ms/step - loss: 1.1573 - accuracy: 0.5973 - val_loss: 1.2215 - val_accuracy: 0.5841
Epoch 85/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1788 - accuracy: 0.5926 - val_loss: 1.2400 - val_accuracy: 0.5701
Epoch 86/200
128/128 [==============================] - 6s 46ms/step - loss: 1.1695 - accuracy: 0.6010 - val_loss: 1.2790 - val_accuracy: 0.5794
Epoch 87/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1658 - accuracy: 0.5955 - val_loss: 1.0967 - val_accuracy: 0.6145
Epoch 88/200
128/128 [==============================] - 6s 43ms/step - loss: 1.1587 - accuracy: 0.5952 - val_loss: 1.1647 - val_accuracy: 0.6028
Epoch 89/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1639 - accuracy: 0.5952 - val_loss: 1.1898 - val_accuracy: 0.6005
Epoch 90/200
128/128 [==============================] - 6s 48ms/step - loss: 1.1914 - accuracy: 0.6010 - val_loss: 1.1508 - val_accuracy: 0.5958
Epoch 91/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1645 - accuracy: 0.5937 - val_loss: 1.6743 - val_accuracy: 0.4556
Epoch 92/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1433 - accuracy: 0.6078 - val_loss: 1.2063 - val_accuracy: 0.5911
Epoch 93/200
128/128 [==============================] - 6s 43ms/step - loss: 1.1270 - accuracy: 0.6096 - val_loss: 1.3283 - val_accuracy: 0.5397
Epoch 94/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1272 - accuracy: 0.6083 - val_loss: 1.3969 - val_accuracy: 0.5561
Epoch 95/200
128/128 [==============================] - 6s 46ms/step - loss: 1.1614 - accuracy: 0.5974 - val_loss: 1.1440 - val_accuracy: 0.6145
Epoch 96/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1465 - accuracy: 0.6020 - val_loss: 1.4556 - val_accuracy: 0.5187
Epoch 97/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1359 - accuracy: 0.6036 - val_loss: 1.1599 - val_accuracy: 0.5818
Epoch 98/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1438 - accuracy: 0.5986 - val_loss: 1.2106 - val_accuracy: 0.5607
Epoch 99/200
128/128 [==============================] - 6s 43ms/step - loss: 1.1534 - accuracy: 0.6015 - val_loss: 1.2408 - val_accuracy: 0.5771
Epoch 100/200
128/128 [==============================] - 6s 47ms/step - loss: 1.1384 - accuracy: 0.6091 - val_loss: 1.3613 - val_accuracy: 0.5678
Epoch 101/200
128/128 [==============================] - 6s 49ms/step - loss: 1.1182 - accuracy: 0.6070 - val_loss: 1.6120 - val_accuracy: 0.4790
Epoch 102/200
128/128 [==============================] - 6s 48ms/step - loss: 1.1714 - accuracy: 0.5882 - val_loss: 1.2130 - val_accuracy: 0.5841
Epoch 103/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1250 - accuracy: 0.6031 - val_loss: 1.5183 - val_accuracy: 0.5327
Epoch 104/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1238 - accuracy: 0.6083 - val_loss: 1.2137 - val_accuracy: 0.5888
Epoch 105/200
128/128 [==============================] - 6s 46ms/step - loss: 1.1159 - accuracy: 0.6167 - val_loss: 1.1614 - val_accuracy: 0.5864
Epoch 106/200
128/128 [==============================] - 6s 45ms/step - loss: 1.1073 - accuracy: 0.6102 - val_loss: 1.1547 - val_accuracy: 0.5935
Epoch 107/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1385 - accuracy: 0.6089 - val_loss: 1.1268 - val_accuracy: 0.6098
Epoch 108/200
128/128 [==============================] - 6s 43ms/step - loss: 1.1352 - accuracy: 0.6039 - val_loss: 1.1905 - val_accuracy: 0.6075
Epoch 109/200
128/128 [==============================] - 6s 45ms/step - loss: 1.1491 - accuracy: 0.6023 - val_loss: 1.1540 - val_accuracy: 0.6051
Epoch 110/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1262 - accuracy: 0.6020 - val_loss: 1.1877 - val_accuracy: 0.6028
Epoch 111/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1356 - accuracy: 0.6013 - val_loss: 1.0888 - val_accuracy: 0.6121
Epoch 112/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1292 - accuracy: 0.6104 - val_loss: 1.1215 - val_accuracy: 0.6121
Epoch 113/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1065 - accuracy: 0.6157 - val_loss: 1.1904 - val_accuracy: 0.5958
Epoch 114/200
128/128 [==============================] - 6s 45ms/step - loss: 1.1245 - accuracy: 0.6047 - val_loss: 1.2435 - val_accuracy: 0.5701
Epoch 115/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1475 - accuracy: 0.6083 - val_loss: 1.2682 - val_accuracy: 0.5561
Epoch 116/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1175 - accuracy: 0.6128 - val_loss: 1.1009 - val_accuracy: 0.6192
Epoch 117/200
128/128 [==============================] - 6s 45ms/step - loss: 1.1111 - accuracy: 0.6083 - val_loss: 1.0995 - val_accuracy: 0.6168
Epoch 118/200
128/128 [==============================] - 6s 45ms/step - loss: 1.1131 - accuracy: 0.6154 - val_loss: 1.7861 - val_accuracy: 0.4322
Epoch 119/200
128/128 [==============================] - 6s 43ms/step - loss: 1.1307 - accuracy: 0.6062 - val_loss: 1.1673 - val_accuracy: 0.6192
Epoch 120/200
128/128 [==============================] - 6s 45ms/step - loss: 1.1191 - accuracy: 0.6107 - val_loss: 1.1400 - val_accuracy: 0.6121
Epoch 121/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1130 - accuracy: 0.6146 - val_loss: 1.4964 - val_accuracy: 0.5304
Epoch 122/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1051 - accuracy: 0.6146 - val_loss: 1.0953 - val_accuracy: 0.6168
Epoch 123/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1026 - accuracy: 0.6175 - val_loss: 1.5946 - val_accuracy: 0.4883
Epoch 124/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1064 - accuracy: 0.6047 - val_loss: 1.1647 - val_accuracy: 0.5958
Epoch 125/200
128/128 [==============================] - 6s 46ms/step - loss: 1.1086 - accuracy: 0.6199 - val_loss: 1.0967 - val_accuracy: 0.6238
Epoch 126/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0806 - accuracy: 0.6259 - val_loss: 1.2440 - val_accuracy: 0.5864
Epoch 127/200
128/128 [==============================] - 6s 45ms/step - loss: 1.1044 - accuracy: 0.6188 - val_loss: 1.1614 - val_accuracy: 0.5818
Epoch 128/200
128/128 [==============================] - 6s 47ms/step - loss: 1.1158 - accuracy: 0.6081 - val_loss: 1.8969 - val_accuracy: 0.3925
Epoch 129/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0897 - accuracy: 0.6209 - val_loss: 1.2186 - val_accuracy: 0.5724
Epoch 130/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0873 - accuracy: 0.6248 - val_loss: 1.0497 - val_accuracy: 0.6519
Epoch 131/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0994 - accuracy: 0.6159 - val_loss: 1.1550 - val_accuracy: 0.6121
Epoch 132/200
128/128 [==============================] - 7s 57ms/step - loss: 1.0992 - accuracy: 0.6259 - val_loss: 1.1689 - val_accuracy: 0.6028
Epoch 133/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1059 - accuracy: 0.6212 - val_loss: 1.2221 - val_accuracy: 0.5841
Epoch 134/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1005 - accuracy: 0.6107 - val_loss: 1.1797 - val_accuracy: 0.5981
Epoch 135/200
128/128 [==============================] - 6s 44ms/step - loss: 1.1099 - accuracy: 0.6162 - val_loss: 1.0848 - val_accuracy: 0.6168
Epoch 136/200
128/128 [==============================] - 6s 47ms/step - loss: 1.0951 - accuracy: 0.6183 - val_loss: 1.2634 - val_accuracy: 0.5561
Epoch 137/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0780 - accuracy: 0.6285 - val_loss: 1.3484 - val_accuracy: 0.5701
Epoch 138/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0966 - accuracy: 0.6196 - val_loss: 1.0993 - val_accuracy: 0.6145
Epoch 139/200
128/128 [==============================] - 6s 45ms/step - loss: 1.1026 - accuracy: 0.6175 - val_loss: 1.1101 - val_accuracy: 0.6192
Epoch 140/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0944 - accuracy: 0.6230 - val_loss: 1.5045 - val_accuracy: 0.5280
Epoch 141/200
128/128 [==============================] - 6s 43ms/step - loss: 1.0699 - accuracy: 0.6251 - val_loss: 1.0447 - val_accuracy: 0.6332
Epoch 142/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0665 - accuracy: 0.6246 - val_loss: 1.0290 - val_accuracy: 0.6519
Epoch 143/200
128/128 [==============================] - 6s 46ms/step - loss: 1.0555 - accuracy: 0.6356 - val_loss: 1.1737 - val_accuracy: 0.5864
Epoch 144/200
128/128 [==============================] - 6s 45ms/step - loss: 1.1110 - accuracy: 0.6073 - val_loss: 1.1600 - val_accuracy: 0.5911
Epoch 145/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0796 - accuracy: 0.6254 - val_loss: 1.0787 - val_accuracy: 0.6425
Epoch 146/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0710 - accuracy: 0.6204 - val_loss: 1.0862 - val_accuracy: 0.6472
Epoch 147/200
128/128 [==============================] - 6s 46ms/step - loss: 1.0682 - accuracy: 0.6369 - val_loss: 1.0001 - val_accuracy: 0.6589
Epoch 148/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0573 - accuracy: 0.6314 - val_loss: 1.0529 - val_accuracy: 0.6449
Epoch 149/200
128/128 [==============================] - 6s 46ms/step - loss: 1.0555 - accuracy: 0.6392 - val_loss: 1.2005 - val_accuracy: 0.6051
Epoch 150/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0825 - accuracy: 0.6282 - val_loss: 1.1202 - val_accuracy: 0.6051
Epoch 151/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0741 - accuracy: 0.6269 - val_loss: 1.5573 - val_accuracy: 0.5280
Epoch 152/200
128/128 [==============================] - 6s 46ms/step - loss: 1.0574 - accuracy: 0.6400 - val_loss: 1.1009 - val_accuracy: 0.6238
Epoch 153/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0702 - accuracy: 0.6264 - val_loss: 1.3309 - val_accuracy: 0.5234
Epoch 154/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0797 - accuracy: 0.6227 - val_loss: 1.1801 - val_accuracy: 0.5958
Epoch 155/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0769 - accuracy: 0.6267 - val_loss: 1.2675 - val_accuracy: 0.5771
Epoch 156/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0581 - accuracy: 0.6314 - val_loss: 1.0109 - val_accuracy: 0.6519
Epoch 157/200
128/128 [==============================] - 6s 48ms/step - loss: 1.0633 - accuracy: 0.6306 - val_loss: 1.1450 - val_accuracy: 0.6051
Epoch 158/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0663 - accuracy: 0.6290 - val_loss: 1.0566 - val_accuracy: 0.6285
Epoch 159/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0525 - accuracy: 0.6364 - val_loss: 1.0305 - val_accuracy: 0.6355
Epoch 160/200
128/128 [==============================] - 7s 55ms/step - loss: 1.0566 - accuracy: 0.6316 - val_loss: 1.2587 - val_accuracy: 0.5794
Epoch 161/200
128/128 [==============================] - 6s 47ms/step - loss: 1.0604 - accuracy: 0.6316 - val_loss: 1.2169 - val_accuracy: 0.5911
Epoch 162/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0521 - accuracy: 0.6290 - val_loss: 1.1192 - val_accuracy: 0.6262
Epoch 163/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0508 - accuracy: 0.6351 - val_loss: 1.1037 - val_accuracy: 0.5911
Epoch 164/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0908 - accuracy: 0.6146 - val_loss: 1.1618 - val_accuracy: 0.5911
Epoch 165/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0547 - accuracy: 0.6403 - val_loss: 1.1286 - val_accuracy: 0.6121
Epoch 166/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0462 - accuracy: 0.6382 - val_loss: 1.0785 - val_accuracy: 0.6308
Epoch 167/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0292 - accuracy: 0.6421 - val_loss: 1.0757 - val_accuracy: 0.6332
Epoch 168/200
128/128 [==============================] - 6s 46ms/step - loss: 1.0374 - accuracy: 0.6406 - val_loss: 1.0911 - val_accuracy: 0.6262
Epoch 169/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0804 - accuracy: 0.6251 - val_loss: 1.3624 - val_accuracy: 0.5631
Epoch 170/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0337 - accuracy: 0.6516 - val_loss: 1.1511 - val_accuracy: 0.5794
Epoch 171/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0658 - accuracy: 0.6285 - val_loss: 1.2314 - val_accuracy: 0.5748
Epoch 172/200
128/128 [==============================] - 6s 47ms/step - loss: 1.0421 - accuracy: 0.6351 - val_loss: 1.1225 - val_accuracy: 0.5958
Epoch 173/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0459 - accuracy: 0.6413 - val_loss: 1.3388 - val_accuracy: 0.5514
Epoch 174/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0311 - accuracy: 0.6461 - val_loss: 1.0377 - val_accuracy: 0.6636
Epoch 175/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0390 - accuracy: 0.6335 - val_loss: 1.2818 - val_accuracy: 0.5421
Epoch 176/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0710 - accuracy: 0.6316 - val_loss: 1.4536 - val_accuracy: 0.5280
Epoch 177/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0460 - accuracy: 0.6356 - val_loss: 1.0014 - val_accuracy: 0.6495
Epoch 178/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0273 - accuracy: 0.6461 - val_loss: 1.0170 - val_accuracy: 0.6659
Epoch 179/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0308 - accuracy: 0.6437 - val_loss: 1.1233 - val_accuracy: 0.5981
Epoch 180/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0389 - accuracy: 0.6398 - val_loss: 1.2872 - val_accuracy: 0.5444
Epoch 181/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0069 - accuracy: 0.6510 - val_loss: 1.0674 - val_accuracy: 0.6215
Epoch 182/200
128/128 [==============================] - 6s 46ms/step - loss: 1.0240 - accuracy: 0.6487 - val_loss: 1.0723 - val_accuracy: 0.6308
Epoch 183/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0623 - accuracy: 0.6254 - val_loss: 1.1839 - val_accuracy: 0.5935
Epoch 184/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0102 - accuracy: 0.6539 - val_loss: 1.1701 - val_accuracy: 0.6075
Epoch 185/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0312 - accuracy: 0.6382 - val_loss: 1.2401 - val_accuracy: 0.5748
Epoch 186/200
128/128 [==============================] - 6s 47ms/step - loss: 1.0352 - accuracy: 0.6479 - val_loss: 1.0852 - val_accuracy: 0.6192
Epoch 187/200
128/128 [==============================] - 6s 49ms/step - loss: 1.0295 - accuracy: 0.6437 - val_loss: 1.1905 - val_accuracy: 0.5888
Epoch 188/200
128/128 [==============================] - 6s 46ms/step - loss: 1.0270 - accuracy: 0.6479 - val_loss: 1.1650 - val_accuracy: 0.6028
Epoch 189/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0203 - accuracy: 0.6450 - val_loss: 1.0069 - val_accuracy: 0.6472
Epoch 190/200
128/128 [==============================] - 6s 46ms/step - loss: 1.0311 - accuracy: 0.6419 - val_loss: 1.2137 - val_accuracy: 0.5818
Epoch 191/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0326 - accuracy: 0.6447 - val_loss: 1.2521 - val_accuracy: 0.5654
Epoch 192/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0392 - accuracy: 0.6419 - val_loss: 0.9843 - val_accuracy: 0.6379
Epoch 193/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0247 - accuracy: 0.6400 - val_loss: 1.0387 - val_accuracy: 0.6262
Epoch 194/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0320 - accuracy: 0.6377 - val_loss: 1.1779 - val_accuracy: 0.6051
Epoch 195/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0279 - accuracy: 0.6427 - val_loss: 1.2350 - val_accuracy: 0.5888
Epoch 196/200
128/128 [==============================] - 6s 46ms/step - loss: 1.0291 - accuracy: 0.6447 - val_loss: 1.0296 - val_accuracy: 0.6565
Epoch 197/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0225 - accuracy: 0.6508 - val_loss: 1.0157 - val_accuracy: 0.6519
Epoch 198/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0251 - accuracy: 0.6461 - val_loss: 1.0513 - val_accuracy: 0.6425
Epoch 199/200
128/128 [==============================] - 6s 44ms/step - loss: 1.0407 - accuracy: 0.6369 - val_loss: 1.0229 - val_accuracy: 0.6449
Epoch 200/200
128/128 [==============================] - 6s 45ms/step - loss: 1.0093 - accuracy: 0.6576 - val_loss: 1.2933 - val_accuracy: 0.5748

Model Evaluation

plt.plot(history_vgg16.history['accuracy'])
plt.plot(history_vgg16.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()

Observations:

• We can see from the above plot that the training accuracy of the model and the validation accuracy are good, but the scores are very less when compared with other previous models
• The model seems to be having a good fit on the data but the data seems be fitting with variations

Model Evaluation on Training dataset

new_model.evaluate(X_train_normalized, y_train_encoded, verbose=1)
model3_TrainDF = model_performance_classification_sklearn_with_threshold(
    new_model, X_train_normalized, y_train_encoded,
)
model3_TrainDF

121/121 [==============================] - 2s 19ms/step - loss: 1.1144 - accuracy: 0.6241
121/121 [==============================] - 2s 16ms/step

	Accuracy	Recall	Precision	F1
0	0.624123	0.624123	0.646204	0.608608

Observations

• The Training accuracy of this model is 62.4% and the other metrics also are in the same range
• We can observe that most of the classes are not classified correctly and requires to be tuned further

Model Evaluation on Validation dataset

new_model.evaluate(X_val_normalized, y_val_encoded, verbose=1)
model3_ValDF = model_performance_classification_sklearn_with_threshold(
    new_model, X_val_normalized,y_val_encoded,
)
model3_ValDF

14/14 [==============================] - 0s 18ms/step - loss: 1.2933 - accuracy: 0.5748
14/14 [==============================] - 0s 16ms/step

	Accuracy	Recall	Precision	F1
0	0.574766	0.574766	0.580763	0.547436

Observations

• The Validation accuracy of this model is 57.5% and the other metrics also are in the same range. Precision has a higher metric
• When compared with the Training accuracy, there is ~5% variance in the Accuracy scores. There is room for improvement to reduce this variance

Model Evaluation on Test dataset

new_model.evaluate(X_test_normalized, y_test_encoded, verbose=2)
model3_TestDF = model_performance_classification_sklearn_with_threshold(new_model, X_test_normalized, y_test_encoded)
model3_TestDF

15/15 - 1s - loss: 1.3466 - accuracy: 0.5579 - 686ms/epoch - 46ms/step
15/15 [==============================] - 0s 16ms/step

	Accuracy	Recall	Precision	F1
0	0.557895	0.557895	0.580586	0.543705

Observations

• The Testing accuracy of this model is 55.8% and the other metrics also are in the same range. Precision has a higher metric
• When compared with the Training accuracy, there is ~7% variance in the Accuracy scores. There is room for improvement to reduce this variance
• The scores are also very less and requires tuning the model further to improve the accuracy

Visualizing the Predictions:

# Visualizing the predicted and correct label of images from test data
rows = 1                                                                        
cols = 5                                                                        
for i in range(cols):
    for j in range(rows):
      random_index = np.random.randint(0, 400)    
      plt.imshow(X_test[random_index])
      plt.show()

      print('Predicted Label', enc.inverse_transform(new_model.predict((X_test_normalized[random_index].reshape(1,64,64,3)))))   
      print('True Label', enc.inverse_transform(y_test_encoded)[random_index])     

1/1 [==============================] - 0s 250ms/step
Predicted Label ['Common wheat']
True Label Common wheat

1/1 [==============================] - 0s 22ms/step
Predicted Label ['Maize']
True Label Maize

1/1 [==============================] - 0s 20ms/step
Predicted Label ['Scentless Mayweed']
True Label Scentless Mayweed

1/1 [==============================] - 0s 20ms/step
Predicted Label ['Charlock']
True Label Small-flowered Cranesbill

1/1 [==============================] - 0s 21ms/step
Predicted Label ['Scentless Mayweed']
True Label Shepherds Purse

Inference:

• We do see that there are images incorrectly classified. Considering the accuracy scores of the training & testing were low, its expected that some of the predictions might be incorrect.

Conclusion

# training performance comparison

models_train_comp_df = pd.concat(
    [
        model0_TrainDF.T,
        model1_TrainDF.T,
        model2_TrainDF.T,
        model3_TrainDF.T,
    ],
    axis=1,
)

models_train_comp_df.columns = [
    "Model 0 - Simple CNN",
    "Model 1 - CNN with Hypertuning",
    "Model 2 - CNN with Data Augmentation",
    "Model 3 - CNN with Transfer Learning",
]

print("Training performance comparison:")
print(models_train_comp_df.T)

# Testing performance comparison

models_test_comp_df = pd.concat(
    [
        model0_TestDF.T,
        model1_TestDF.T,
        model2_TestDF.T,
        model3_TestDF.T,
    ],
    axis=1,
)
models_test_comp_df.columns = [
    "Model 0 - Simple CNN",
    "Model 1 - CNN with Hypertuning",
    "Model 2 - CNN with Data Augmentation",
    "Model 3 - CNN with Transfer Learning",
]

print("\n\n")
print("Testing performance comparison:")
print(models_test_comp_df.T)

Training performance comparison:
                                      Accuracy    Recall  Precision        F1
Model 0 - Simple CNN                  1.000000  1.000000   1.000000  1.000000
Model 1 - CNN with Hypertuning        0.973746  0.973746   0.974874  0.973941
Model 2 - CNN with Data Augmentation  0.946452  0.946452   0.947343  0.946507
Model 3 - CNN with Transfer Learning  0.624123  0.624123   0.646204  0.608608



Testing performance comparison:
                                      Accuracy    Recall  Precision        F1
Model 0 - Simple CNN                  0.732632  0.732632   0.729495  0.727110
Model 1 - CNN with Hypertuning        0.886316  0.886316   0.894756  0.887596
Model 2 - CNN with Data Augmentation  0.911579  0.911579   0.914286  0.912290
Model 3 - CNN with Transfer Learning  0.557895  0.557895   0.580586  0.543705

Inferences:

• We can observe from the above comparison of the various performance metrics, the Model 2 (CNN with Data Augmentation) is the best model because it has the highest accuracy of Training (94.6%) & Testing (91.2%) data without overfitting and also predicted the majority of the classes better than the other models as illustratced in the Confusion Matrix
• The Model 1 CNN with Hypertuning has provided a better training accuracy (97.4%) result than the Model 2 with Data Augumentation but the testing accuracy is at 88.6%
• The test accuracy of the Model 3 is only 55.7% and the other metrics Recall, Precision & F1 almost has a similar performance score
• For this business case, the model utilizing the VGG16 - Transfer Learning has not helped in improving the model and achieve a better accuracy
• There is a 9% variance in the accuracy scores in Model 1 when compared with the 3% variance of Model 2. The performance metrics of F1, Precision and Recall are also very good in the testing scores of Model 2

Model 2 has better accuracy variance and chance of predicting the results more accurately. We can use this model to classify the plant species

---